News & Updates
The latest news and updates from companies in the WLTH portfolio.
Anthropic's Claude Fable 5 Just Launched
On June 9, 2026, Anthropic released Claude Fable 5, the first model in a brand new top tier the company calls the Mythos class. This is not another point upgrade in the Opus line. Anthropic describes Fable 5 as a model that sits above Opus in raw capability. It is state of the art on nearly every benchmark it was tested on, and it is the most powerful model the company has ever made available to the general public. It launched alongside a sibling model, Claude Mythos 5. That model is the same underlying system with the safety guardrails lifted, available only to a small set of vetted users. The new tier is the real headline. For two years Anthropic ran a three step ladder of Haiku, Sonnet, and Opus. Mythos is a fourth step above all of them, and Fable 5 is the version of that step the public can actually use. The timing is its own story. Fable 5 arrived only days after Anthropic warned that frontier AI is approaching recursive self improvement and urged the industry to agree on a coordinated brake on development. Releasing your most powerful model yet, days after that warning, struck a lot of observers as either a contradiction or a carefully chosen strategy. In this article we cover what Fable 5 is, how the Mythos class works, the full benchmark picture against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, the safeguard system, pricing, availability, and how the restricted Mythos 5 model fits in. What Claude Fable 5 Actually Is Claude Fable 5 is a frontier reasoning model and the first generally available member of Anthropic's Mythos class. Here are the basics at a glance: * API identifier: , with und on AWS * Same model as Mythos 5, separated only by a safety layer added at inference time * Built for long horizon agentic work, not quick single turn chat * State of the art vision, including reading charts and rebuilding code from screenshots * 1 million token context window, per Artificial Analysis The most important fact is that Fable 5 and Mythos 5 are the same model. The difference is not the weights or the training. It is the safety layer wrapped around Fable 5 that intercepts high risk requests. So every benchmark number Fable 5 posts is the Mythos class capability operating with a safety net underneath it. Anthropic positions the model for autonomous work that runs for hours or even days inside an agent harness. In that setting the model plans a multi step job, calls tools, reads the results, validates its own output, and corrects course without a human in the loop. The trait partners keep citing is self verification. Rakuten told Anthropic the model reflects on and validates its own work, which is what makes the autonomous operation practical rather than risky. On context size, the picture firmed up after launch. Independent benchmarking site Artificial Analysis lists a 1 million token window, matching the figure that circulated in early summaries. The companion 128,000 token maximum output number is still not confirmed in Anthropic's own materials, so treat the output ceiling as unverified for now. The Mythos Class Explained Anthropic's lineup has always been a ladder: * Haiku is the small, fast, cheap tier * Sonnet is the balanced workhorse * Opus is the frontier reasoning tier * Mythos is the new step above all three Mythos represents a capability level Anthropic had been holding back from general release, because of the risks a model that strong could pose in the wrong hands. Fable 5 is what you get when Anthropic takes a Mythos class model and adds the guardrails needed to make it safe for a broad audience. The pitch is that the safeguards are not a tax on the model. They are the thing that makes shipping it possible at all. That framing matters for the price. You are not paying twice the cost of Opus 4.8 for a slightly better Opus. You are paying for a step up to a new tier, delivered with a safety system the older tiers never needed. A Brief History of Mythos and Project Glasswing Mythos did not appear out of nowhere on launch day. The first Mythos model, Claude Mythos Preview, shipped quietly in April 2026 through a limited program named Project Glasswing. Project Glasswing was not a public product. It was a controlled access program aimed at a narrow set of users, mainly cyber defenders and critical infrastructure providers. These are organizations that could use a frontier model's offensive security knowledge to strengthen their own defenses. Anthropic used the program to learn how a Mythos class model behaves in the field before deciding whether to release it more widely. The June 9 launch is the graduation of that experiment into two official products: * Claude Mythos 5, the direct upgrade for Glasswing partners and select biology researchers, available only through trusted access * Claude Fable 5, the safeguarded version built for everyone else Anthropic has also signaled that the trusted access program for vetted cybersecurity organizations will broaden, and that enrollment for a biology research program is beginning. Benchmark Results Anthropic published a full benchmark table with the launch, putting the Mythos class against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Launch coverage from VentureBeat und Tom's Hardware echoed the topline that it is state of the art on nearly all of them. Two things matter for reading the table. The Anthropic column combines Mythos 5 and Fable 5, showing the higher of the two, which are normally within 1 to 3 points of each other. The starred rows are the exception. On those, Fable 5's safeguards route requests to Opus 4.8, so the public Fable 5 score lands closer to Opus 4.8 while the starred figure reflects the unrestricted Mythos 5. Starred rows are where Fable 5's safeguards trigger a fallback. The number shown is Mythos 5, and public Fable 5 performs closer to Opus 4.8 on those tasks. On Terminal Bench 2.1, GPT-5.5's 83.4% is via Codex CLI and Gemini's 70.7% via Gemini CLI. The standouts are the unstarred rows, where Fable 5 and Mythos 5 are effectively the same. Fable 5 leads every rival on agentic coding, knowledge work, vision, tool use, computer use, and legal reasoning. The SWE-Bench Pro gap is the clearest signal, with Fable 5 at 80.3% sitting more than 11 points above Opus 4.8 and nearly 22 above GPT-5.5, a margin that compounds across every step of a long autonomous job. On the hardest FrontierCode Diamond split it more than doubles Opus 4.8 and roughly quintuples GPT-5.5. A few results outside the official table round out the picture: * Analytics: first model to score 90% on Hex's benchmark of complex, long running analytical tasks * Finance: highest score of any model on Hebbia's senior level finance benchmark * Memory: a persistent file based memory task improved Fable 5's performance three times more than Opus 4.8 * Spreadsheets: beats Opus 4.8 at every effort level while finishing runs 25 to 30% faster How Fable 5 Compares to Opus 4.8, GPT-5.5, and Gemini 3.1 Pro Stripping the benchmarks down to positioning, the four models line up like this. The pattern from the benchmark table is consistent. Fable 5 leads on the unsafeguarded categories, with Opus 4.8 as the closest follower. GPT-5.5 sits a step behind on coding but stays competitive on vision and spatial reasoning, and Gemini 3.1 Pro trails across the board. The tradeoff is price and the safety layer. Fable 5 costs twice what Opus 4.8 does and routes certain requests away from itself, which none of the other models here do. Independent Benchmarks from Artificial Analysis The first independent confirmation came from Artificial Analysis, which folded Fable 5 into its public leaderboard within a day of launch. Fable 5 debuted at the very top. The headline numbers: * Intelligence Index: 65, ranked number one, against a roughly 36 average for comparable models * GDPval-AA (agentic real world work): 1,932, number one, with Anthropic models taking three of the top four spots * Coding and agentic sub scores: 62 and 80.7, both at or near the top * Fallback rate: just 2% of GDPval-AA tasks deferred to Opus 4.8, matching Anthropic's under 5% claim The independent data also surfaced weaknesses the partner testimonials gloss over. The biggest one is speed: * Output speed: 60.3 tokens per second, ranked 72nd of 152 models, squarely mid pack * Time to first token: around 82 seconds, far above the peer median of under 3 seconds That latency is a direct consequence of the heavy chain of thought reasoning the model runs before answering. This is a model built for deep, long horizon work, not snappy back and forth chat. Cost is the other caveat. Artificial Analysis lists Fable 5 at the expensive end of its chart, quoting an input rate of $12.50 per million tokens against the $10 in Anthropic's own materials, with a blended rate around $8.20 once cache hits are factored in. Either way, it is among the priciest models on the board. One scoreboard is still blank. Fable 5 has not yet appeared on LMArena's Chatbot Arena leaderboard, which ranks models by head to head human preference votes. That is expected for a model this new, since Arena needs a large volume of comparisons before assigning a reliable rating. A human preference ranking will be one of the more interesting data points to watch in the coming weeks. Real World Performance from Launch Partners Benchmarks are one signal. The launch partner reports point in the same direction. Stripe produced the most cited result. Anthropic says Fable 5 compressed months of engineering into days. The specific example is a migration across a 50 million line Ruby codebase that the model completed in a single day, work that would otherwise have taken a full engineering team over two months. Rakuten emphasized reliability over raw speed. The company reported that Fable 5 reflects on and validates its own work, letting it run autonomous operations where the model is trusted to check itself rather than handing every step back to a human. Hex contributed the 90% analytics milestone, the first time any model cleared that bar on its suite of complex tasks. AWS framed the model as purpose built for long running, asynchronous execution, the kind of job that can run for days inside a harness before producing a result. The Safeguard System The safeguards are what make Fable 5 a public product, so they deserve a close look. Fable 5 ships with classifiers covering four high risk domains. When a request trips one, the model blocks its own response and falls back to Claude Opus 4.8, the older and more conservative model, to handle the query safely. The domains are: * Cybersecurity, blocking exploitation and offensive cyber tasks * Biology and chemistry, blocking risky dual use research * Distillation, blocking attempts to extract or copy the model's capabilities The key number is how often this happens. Anthropic says more than 95% of Fable 5 sessions involve no fallback at all. Only around 5% hit a classifier and defer to Opus 4.8, so for the overwhelming majority of normal use you are talking to the full Mythos class model. The red team results are the other half of the safety case: * No universal jailbreaks across more than 1,000 hours of external bug bounty testing * Zero harmful single turn completions on cyberattack planning across 30 jailbreak techniques * One external partner rated it the most robust safeguards of any model tested * Mythos 5's misaligned behavior measured low, similar to Opus 4.8 There is a cost dimension too. When a request is routed to Opus 4.8, you pay Opus prices for that portion. Anthropic also requires a mandatory 30 day retention of inputs and outputs for all users, with human review capability, which it frames as a defense against novel attacks. On AWS this is enforced through a setting that must be enabled before the model can be invoked. What Mythos 5 Can Do That Fable 5 Cannot The starred rows in the benchmark table above are exactly where the two models diverge. On those, Mythos 5 answers at full capability while Fable 5's safeguards route the request to Opus 4.8, so the public model performs closer to Opus 4.8 than to the starred score. The gap is widest in the high risk domains: * Cybersecurity (ExploitBench): Mythos 5 captures 78.0% against Opus 4.8 at 40.0% and GPT-5.5 at 34.0%, nearly doubling the previous frontier * Biology hard (BioMysteryBench): Mythos 5 leads at 46.1% versus Opus 4.8 at 40.0% * Health (HealthBench Professional): Mythos 5 at 66.0% versus Opus 4.8 at 56.9% * Humanity's Last Exam with tools: 64.5% for the Mythos class versus 57.9% for Opus 4.8 A general user hitting these topics through Fable 5 is redirected to Opus 4.8, which is why public Fable 5 lands near the Opus column on the starred rows. Mythos 5, available only to vetted cyber defenders and approved biology researchers, is the version that answers in full. This is the core of Anthropic's argument for the whole release. A model this capable in cybersecurity and biology is genuinely dangerous in the wrong hands, so that capability is locked behind trusted access while Fable 5 lets everyone else use the same intelligence for everything that is not high risk. Pricing Claude Fable 5 and Claude Mythos 5 are both priced at $10 per million input tokens and $50 per million output tokens. For context, Claude Opus 4.8 costs $5 and $25, which makes Fable 5 exactly twice the price of the model just below it. Anthropic prefers to compare up rather than down. Against the earlier Claude Mythos Preview, Fable 5 costs less than half as much, so within the Mythos class the price has dropped sharply even as capability went up. Whether the premium is worth it depends on the work. For routine chat, Opus 4.8 or a cheaper tier is the obvious pick. For long autonomous jobs where a higher pass rate compounds across hundreds of steps, like the Stripe migration, the premium can pay for itself by finishing tasks a cheaper model would fail. Availability and Access Anthropic shipped Fable 5 across a wide set of platforms on launch day. * Claude API and Claude Platform, plus consumer and business subscriptions * Amazon Bedrock, live in US East (N. Virginia) and Europe (Stockholm) at launch * GitHub Copilot, generally available * Harvey, the legal AI platform There is a rollout window worth noting. Through June 22, 2026, Fable 5 is included at no extra cost on Pro, Max, Team, and seat based Enterprise plans. From June 23, access shifts to a usage credit model, so the free inclusion is effectively a two week introductory period. Claude Mythos 5 is the exception to all this breadth. It is restricted to Project Glasswing partners and a small set of approved biology researchers, and it is not part of any general subscription or public API tier. The Timing Controversy The release cannot be separated from its context. Just days before launching Fable 5, Anthropic warned that AI systems are advancing toward recursive self improvement, the point at which models can improve themselves without human intervention. The company paired that warning with a call for major AI labs to agree on a coordinated brake on frontier development. Then it shipped its most powerful model yet. The apparent contradiction was not lost on anyone covering the launch. Releasing a Mythos class model to the public, days after arguing that AI is becoming too dangerous, reads as either a reversal or a deliberate strategy. Anthropic's answer is that the safeguards are the entire point. The argument runs that a Mythos class capability is coming whether Anthropic ships it or not, so it is better for the first public model at this tier to arrive with hard guardrails, fallback routing, and mandatory monitoring than to wait for a competitor to release something equally capable with no protections. Whether that holds up is fair to debate. The safeguards are real and the red team numbers are strong, but they depend on classifiers that have to catch every dangerous request, data retention that raises privacy questions, and a fallback model users do not control. The launch is a live test of whether a model this capable can be released responsibly. Where Fable 5 Leaves Questions Open A few things are not fully nailed down yet: * Output ceiling: the 1 million token context is now corroborated by Artificial Analysis, but the rumored 128,000 token output limit is still unverified in Anthropic's materials * Vendor figures: Anthropic's benchmark numbers are its own, and the full cross model comparison across every tested category has not been published * Fallback experience: a model that silently routes 5% of sessions to an older model behaves differently near the boundary of the blocked domains, and the quality of that handoff is untested in public None of this undercuts the headline. On the benchmarks Anthropic did publish, and on Artificial Analysis's independent leaderboard, Fable 5 is the strongest model the company has released to the public. The full picture will sharpen as the model card and more independent evaluations land. Why This Release Matters Three things make this launch significant beyond the benchmark numbers. A new tier, public on day one. For two years the frontier conversation has been about incremental gains within the Opus, GPT, and Gemini flagship lines. Mythos is a deliberate step above that ladder, and Fable 5 makes it generally available rather than locking it to a research preview. A safeguard template the industry will study. Anthropic is betting that the way to ship a dangerous capability is to wrap it in classifiers, fallback routing, red teaming, and monitoring, then release the safe version broadly while the unrestricted version stays behind a trusted access wall. If it works, it becomes the playbook. If it fails, it becomes the cautionary tale. A coding gap big enough to change what agents can do. An 80.3% SWE-Bench Pro pass rate and a Stripe migration finished in a day point to a model that can carry long, complex jobs end to end rather than assisting step by step. That is the difference between a coding assistant and an autonomous engineer. For anyone who wants to see Claude together with other AI models, Fello AI puts Claude, GPT, Gemini, DeepSeek, Perplexity, and more into a single native app for Mac, iPhone, and iPad, so you can test the same task across models and pick the right one for the job without managing a stack of subscriptions.
Anthropic's Claude Fable 5 Just Launched
On June 9, 2026, Anthropic released Claude Fable 5, the first model in a brand new top tier the company calls the Mythos class. This is not another point upgrade in the Opus line. Anthropic describes Fable 5 as a model that sits above Opus in raw capability. It is state of the art on nearly every benchmark it was tested on, and it is the most powerful model the company has ever made available to the general public. It launched alongside a sibling model, Claude Mythos 5. That model is the same underlying system with the safety guardrails lifted, available only to a small set of vetted users. The new tier is the real headline. For two years Anthropic ran a three step ladder of Haiku, Sonnet, and Opus. Mythos is a fourth step above all of them, and Fable 5 is the version of that step the public can actually use. The timing is its own story. Fable 5 arrived only days after Anthropic warned that frontier AI is approaching recursive self improvement and urged the industry to agree on a coordinated brake on development. Releasing your most powerful model yet, days after that warning, struck a lot of observers as either a contradiction or a carefully chosen strategy. In this article we cover what Fable 5 is, how the Mythos class works, the full benchmark picture against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, the safeguard system, pricing, availability, and how the restricted Mythos 5 model fits in. What Claude Fable 5 Actually Is Claude Fable 5 is a frontier reasoning model and the first generally available member of Anthropic's Mythos class. Here are the basics at a glance: * API identifier: , with a on AWS * Same model as Mythos 5, separated only by a safety layer added at inference time * Built for long horizon agentic work, not quick single turn chat * State of the art vision, including reading charts and rebuilding code from screenshots * 1 million token context window, per Artificial Analysis The most important fact is that Fable 5 and Mythos 5 are the same model. The difference is not the weights or the training. It is the safety layer wrapped around Fable 5 that intercepts high risk requests. So every benchmark number Fable 5 posts is the Mythos class capability operating with a safety net underneath it. Anthropic positions the model for autonomous work that runs for hours or even days inside an agent harness. In that setting the model plans a multi step job, calls tools, reads the results, validates its own output, and corrects course without a human in the loop. The trait partners keep citing is self verification. Rakuten told Anthropic the model reflects on and validates its own work, which is what makes the autonomous operation practical rather than risky. On context size, the picture firmed up after launch. Independent benchmarking site Artificial Analysis lists a 1 million token window, matching the figure that circulated in early summaries. The companion 128,000 token maximum output number is still not confirmed in Anthropic's own materials, so treat the output ceiling as unverified for now. The Mythos Class Explained Anthropic's lineup has always been a ladder: * Haiku is the small, fast, cheap tier * Sonnet is the balanced workhorse * Opus is the frontier reasoning tier * Mythos is the new step above all three Mythos represents a capability level Anthropic had been holding back from general release, because of the risks a model that strong could pose in the wrong hands. Fable 5 is what you get when Anthropic takes a Mythos class model and adds the guardrails needed to make it safe for a broad audience. The pitch is that the safeguards are not a tax on the model. They are the thing that makes shipping it possible at all. That framing matters for the price. You are not paying twice the cost of Opus 4.8 for a slightly better Opus. You are paying for a step up to a new tier, delivered with a safety system the older tiers never needed. A Brief History of Mythos and Project Glasswing Mythos did not appear out of nowhere on launch day. The first Mythos model, Claude Mythos Preview, shipped quietly in April 2026 through a limited program named Project Glasswing. Project Glasswing was not a public product. It was a controlled access program aimed at a narrow set of users, mainly cyber defenders and critical infrastructure providers. These are organizations that could use a frontier model's offensive security knowledge to strengthen their own defenses. Anthropic used the program to learn how a Mythos class model behaves in the field before deciding whether to release it more widely. The June 9 launch is the graduation of that experiment into two official products: * Claude Mythos 5, the direct upgrade for Glasswing partners and select biology researchers, available only through trusted access * Claude Fable 5, the safeguarded version built for everyone else Anthropic has also signaled that the trusted access program for vetted cybersecurity organizations will broaden, and that enrollment for a biology research program is beginning. Benchmark Results Anthropic published a full benchmark table with the launch, putting the Mythos class against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Launch coverage from VentureBeat a Tom's Hardware echoed the topline that it is state of the art on nearly all of them. Two things matter for reading the table. The Anthropic column combines Mythos 5 and Fable 5, showing the higher of the two, which are normally within 1 to 3 points of each other. The starred rows are the exception. On those, Fable 5's safeguards route requests to Opus 4.8, so the public Fable 5 score lands closer to Opus 4.8 while the starred figure reflects the unrestricted Mythos 5. Starred rows are where Fable 5's safeguards trigger a fallback. The number shown is Mythos 5, and public Fable 5 performs closer to Opus 4.8 on those tasks. On Terminal Bench 2.1, GPT-5.5's 83.4% is via Codex CLI and Gemini's 70.7% via Gemini CLI. The standouts are the unstarred rows, where Fable 5 and Mythos 5 are effectively the same. Fable 5 leads every rival on agentic coding, knowledge work, vision, tool use, computer use, and legal reasoning. The SWE-Bench Pro gap is the clearest signal, with Fable 5 at 80.3% sitting more than 11 points above Opus 4.8 and nearly 22 above GPT-5.5, a margin that compounds across every step of a long autonomous job. On the hardest FrontierCode Diamond split it more than doubles Opus 4.8 and roughly quintuples GPT-5.5. A few results outside the official table round out the picture: * Analytics: first model to score 90% on Hex's benchmark of complex, long running analytical tasks * Finance: highest score of any model on Hebbia's senior level finance benchmark * Memory: a persistent file based memory task improved Fable 5's performance three times more than Opus 4.8 * Spreadsheets: beats Opus 4.8 at every effort level while finishing runs 25 to 30% faster How Fable 5 Compares to Opus 4.8, GPT-5.5, and Gemini 3.1 Pro Stripping the benchmarks down to positioning, the four models line up like this. The pattern from the benchmark table is consistent. Fable 5 leads on the unsafeguarded categories, with Opus 4.8 as the closest follower. GPT-5.5 sits a step behind on coding but stays competitive on vision and spatial reasoning, and Gemini 3.1 Pro trails across the board. The tradeoff is price and the safety layer. Fable 5 costs twice what Opus 4.8 does and routes certain requests away from itself, which none of the other models here do. Independent Benchmarks from Artificial Analysis The first independent confirmation came from Artificial Analysis, which folded Fable 5 into its public leaderboard within a day of launch. Fable 5 debuted at the very top. The headline numbers: * Intelligence Index: 65, ranked number one, against a roughly 36 average for comparable models * GDPval-AA (agentic real world work): 1,932, number one, with Anthropic models taking three of the top four spots * Coding and agentic sub scores: 62 and 80.7, both at or near the top * Fallback rate: just 2% of GDPval-AA tasks deferred to Opus 4.8, matching Anthropic's under 5% claim The independent data also surfaced weaknesses the partner testimonials gloss over. The biggest one is speed: * Output speed: 60.3 tokens per second, ranked 72nd of 152 models, squarely mid pack * Time to first token: around 82 seconds, far above the peer median of under 3 seconds That latency is a direct consequence of the heavy chain of thought reasoning the model runs before answering. This is a model built for deep, long horizon work, not snappy back and forth chat. Cost is the other caveat. Artificial Analysis lists Fable 5 at the expensive end of its chart, quoting an input rate of $12.50 per million tokens against the $10 in Anthropic's own materials, with a blended rate around $8.20 once cache hits are factored in. Either way, it is among the priciest models on the board. One scoreboard is still blank. Fable 5 has not yet appeared on LMArena's Chatbot Arena leaderboard, which ranks models by head to head human preference votes. That is expected for a model this new, since Arena needs a large volume of comparisons before assigning a reliable rating. A human preference ranking will be one of the more interesting data points to watch in the coming weeks. Real World Performance from Launch Partners Benchmarks are one signal. The launch partner reports point in the same direction. Stripe produced the most cited result. Anthropic says Fable 5 compressed months of engineering into days. The specific example is a migration across a 50 million line Ruby codebase that the model completed in a single day, work that would otherwise have taken a full engineering team over two months. Rakuten emphasized reliability over raw speed. The company reported that Fable 5 reflects on and validates its own work, letting it run autonomous operations where the model is trusted to check itself rather than handing every step back to a human. Hex contributed the 90% analytics milestone, the first time any model cleared that bar on its suite of complex tasks. AWS framed the model as purpose built for long running, asynchronous execution, the kind of job that can run for days inside a harness before producing a result. The Safeguard System The safeguards are what make Fable 5 a public product, so they deserve a close look. Fable 5 ships with classifiers covering four high risk domains. When a request trips one, the model blocks its own response and falls back to Claude Opus 4.8, the older and more conservative model, to handle the query safely. The domains are: * Cybersecurity, blocking exploitation and offensive cyber tasks * Biology and chemistry, blocking risky dual use research * Distillation, blocking attempts to extract or copy the model's capabilities The key number is how often this happens. Anthropic says more than 95% of Fable 5 sessions involve no fallback at all. Only around 5% hit a classifier and defer to Opus 4.8, so for the overwhelming majority of normal use you are talking to the full Mythos class model. The red team results are the other half of the safety case: * No universal jailbreaks across more than 1,000 hours of external bug bounty testing * Zero harmful single turn completions on cyberattack planning across 30 jailbreak techniques * One external partner rated it the most robust safeguards of any model tested * Mythos 5's misaligned behavior measured low, similar to Opus 4.8 There is a cost dimension too. When a request is routed to Opus 4.8, you pay Opus prices for that portion. Anthropic also requires a mandatory 30 day retention of inputs and outputs for all users, with human review capability, which it frames as a defense against novel attacks. On AWS this is enforced through a setting that must be enabled before the model can be invoked. What Mythos 5 Can Do That Fable 5 Cannot The starred rows in the benchmark table above are exactly where the two models diverge. On those, Mythos 5 answers at full capability while Fable 5's safeguards route the request to Opus 4.8, so the public model performs closer to Opus 4.8 than to the starred score. The gap is widest in the high risk domains: * Cybersecurity (ExploitBench): Mythos 5 captures 78.0% against Opus 4.8 at 40.0% and GPT-5.5 at 34.0%, nearly doubling the previous frontier * Biology hard (BioMysteryBench): Mythos 5 leads at 46.1% versus Opus 4.8 at 40.0% * Health (HealthBench Professional): Mythos 5 at 66.0% versus Opus 4.8 at 56.9% * Humanity's Last Exam with tools: 64.5% for the Mythos class versus 57.9% for Opus 4.8 A general user hitting these topics through Fable 5 is redirected to Opus 4.8, which is why public Fable 5 lands near the Opus column on the starred rows. Mythos 5, available only to vetted cyber defenders and approved biology researchers, is the version that answers in full. This is the core of Anthropic's argument for the whole release. A model this capable in cybersecurity and biology is genuinely dangerous in the wrong hands, so that capability is locked behind trusted access while Fable 5 lets everyone else use the same intelligence for everything that is not high risk. Pricing Claude Fable 5 and Claude Mythos 5 are both priced at $10 per million input tokens and $50 per million output tokens. For context, Claude Opus 4.8 costs $5 and $25, which makes Fable 5 exactly twice the price of the model just below it. Anthropic prefers to compare up rather than down. Against the earlier Claude Mythos Preview, Fable 5 costs less than half as much, so within the Mythos class the price has dropped sharply even as capability went up. Whether the premium is worth it depends on the work. For routine chat, Opus 4.8 or a cheaper tier is the obvious pick. For long autonomous jobs where a higher pass rate compounds across hundreds of steps, like the Stripe migration, the premium can pay for itself by finishing tasks a cheaper model would fail. Availability and Access Anthropic shipped Fable 5 across a wide set of platforms on launch day. * Claude API and Claude Platform, plus consumer and business subscriptions * Amazon Bedrock, live in US East (N. Virginia) and Europe (Stockholm) at launch * GitHub Copilot, generally available * Harvey, the legal AI platform There is a rollout window worth noting. Through June 22, 2026, Fable 5 is included at no extra cost on Pro, Max, Team, and seat based Enterprise plans. From June 23, access shifts to a usage credit model, so the free inclusion is effectively a two week introductory period. Claude Mythos 5 is the exception to all this breadth. It is restricted to Project Glasswing partners and a small set of approved biology researchers, and it is not part of any general subscription or public API tier. The Timing Controversy The release cannot be separated from its context. Just days before launching Fable 5, Anthropic warned that AI systems are advancing toward recursive self improvement, the point at which models can improve themselves without human intervention. The company paired that warning with a call for major AI labs to agree on a coordinated brake on frontier development. Then it shipped its most powerful model yet. The apparent contradiction was not lost on anyone covering the launch. Releasing a Mythos class model to the public, days after arguing that AI is becoming too dangerous, reads as either a reversal or a deliberate strategy. Anthropic's answer is that the safeguards are the entire point. The argument runs that a Mythos class capability is coming whether Anthropic ships it or not, so it is better for the first public model at this tier to arrive with hard guardrails, fallback routing, and mandatory monitoring than to wait for a competitor to release something equally capable with no protections. Whether that holds up is fair to debate. The safeguards are real and the red team numbers are strong, but they depend on classifiers that have to catch every dangerous request, data retention that raises privacy questions, and a fallback model users do not control. The launch is a live test of whether a model this capable can be released responsibly. Where Fable 5 Leaves Questions Open A few things are not fully nailed down yet: * Output ceiling: the 1 million token context is now corroborated by Artificial Analysis, but the rumored 128,000 token output limit is still unverified in Anthropic's materials * Vendor figures: Anthropic's benchmark numbers are its own, and the full cross model comparison across every tested category has not been published * Fallback experience: a model that silently routes 5% of sessions to an older model behaves differently near the boundary of the blocked domains, and the quality of that handoff is untested in public None of this undercuts the headline. On the benchmarks Anthropic did publish, and on Artificial Analysis's independent leaderboard, Fable 5 is the strongest model the company has released to the public. The full picture will sharpen as the model card and more independent evaluations land. Why This Release Matters Three things make this launch significant beyond the benchmark numbers. A new tier, public on day one. For two years the frontier conversation has been about incremental gains within the Opus, GPT, and Gemini flagship lines. Mythos is a deliberate step above that ladder, and Fable 5 makes it generally available rather than locking it to a research preview. A safeguard template the industry will study. Anthropic is betting that the way to ship a dangerous capability is to wrap it in classifiers, fallback routing, red teaming, and monitoring, then release the safe version broadly while the unrestricted version stays behind a trusted access wall. If it works, it becomes the playbook. If it fails, it becomes the cautionary tale. A coding gap big enough to change what agents can do. An 80.3% SWE-Bench Pro pass rate and a Stripe migration finished in a day point to a model that can carry long, complex jobs end to end rather than assisting step by step. That is the difference between a coding assistant and an autonomous engineer. For anyone who wants to see Claude together with other AI models, Fello AI puts Claude, GPT, Gemini, DeepSeek, Perplexity, and more into a single native app for Mac, iPhone, and iPad, so you can test the same task across models and pick the right one for the job without managing a stack of subscriptions.
Anthropic's Claude Fable 5 Just Launched
On June 9, 2026, Anthropic released Claude Fable 5, the first model in a brand new top tier the company calls the Mythos class. This is not another point upgrade in the Opus line. Anthropic describes Fable 5 as a model that sits above Opus in raw capability. It is state of the art on nearly every benchmark it was tested on, and it is the most powerful model the company has ever made available to the general public. It launched alongside a sibling model, Claude Mythos 5. That model is the same underlying system with the safety guardrails lifted, available only to a small set of vetted users. The new tier is the real headline. For two years Anthropic ran a three step ladder of Haiku, Sonnet, and Opus. Mythos is a fourth step above all of them, and Fable 5 is the version of that step the public can actually use. The timing is its own story. Fable 5 arrived only days after Anthropic warned that frontier AI is approaching recursive self improvement and urged the industry to agree on a coordinated brake on development. Releasing your most powerful model yet, days after that warning, struck a lot of observers as either a contradiction or a carefully chosen strategy. In this article we cover what Fable 5 is, how the Mythos class works, the full benchmark picture against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, the safeguard system, pricing, availability, and how the restricted Mythos 5 model fits in. What Claude Fable 5 Actually Is Claude Fable 5 is a frontier reasoning model and the first generally available member of Anthropic's Mythos class. Here are the basics at a glance: * API identifier: , with and on AWS * Same model as Mythos 5, separated only by a safety layer added at inference time * Built for long horizon agentic work, not quick single turn chat * State of the art vision, including reading charts and rebuilding code from screenshots * 1 million token context window, per Artificial Analysis The most important fact is that Fable 5 and Mythos 5 are the same model. The difference is not the weights or the training. It is the safety layer wrapped around Fable 5 that intercepts high risk requests. So every benchmark number Fable 5 posts is the Mythos class capability operating with a safety net underneath it. Anthropic positions the model for autonomous work that runs for hours or even days inside an agent harness. In that setting the model plans a multi step job, calls tools, reads the results, validates its own output, and corrects course without a human in the loop. The trait partners keep citing is self verification. Rakuten told Anthropic the model reflects on and validates its own work, which is what makes the autonomous operation practical rather than risky. On context size, the picture firmed up after launch. Independent benchmarking site Artificial Analysis lists a 1 million token window, matching the figure that circulated in early summaries. The companion 128,000 token maximum output number is still not confirmed in Anthropic's own materials, so treat the output ceiling as unverified for now. The Mythos Class Explained Anthropic's lineup has always been a ladder: * Haiku is the small, fast, cheap tier * Sonnet is the balanced workhorse * Opus is the frontier reasoning tier * Mythos is the new step above all three Mythos represents a capability level Anthropic had been holding back from general release, because of the risks a model that strong could pose in the wrong hands. Fable 5 is what you get when Anthropic takes a Mythos class model and adds the guardrails needed to make it safe for a broad audience. The pitch is that the safeguards are not a tax on the model. They are the thing that makes shipping it possible at all. That framing matters for the price. You are not paying twice the cost of Opus 4.8 for a slightly better Opus. You are paying for a step up to a new tier, delivered with a safety system the older tiers never needed. A Brief History of Mythos and Project Glasswing Mythos did not appear out of nowhere on launch day. The first Mythos model, Claude Mythos Preview, shipped quietly in April 2026 through a limited program named Project Glasswing. Project Glasswing was not a public product. It was a controlled access program aimed at a narrow set of users, mainly cyber defenders and critical infrastructure providers. These are organizations that could use a frontier model's offensive security knowledge to strengthen their own defenses. Anthropic used the program to learn how a Mythos class model behaves in the field before deciding whether to release it more widely. The June 9 launch is the graduation of that experiment into two official products: * Claude Mythos 5, the direct upgrade for Glasswing partners and select biology researchers, available only through trusted access * Claude Fable 5, the safeguarded version built for everyone else Anthropic has also signaled that the trusted access program for vetted cybersecurity organizations will broaden, and that enrollment for a biology research program is beginning. Benchmark Results Anthropic published a full benchmark table with the launch, putting the Mythos class against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Launch coverage from VentureBeat and Tom's Hardware echoed the topline that it is state of the art on nearly all of them. Two things matter for reading the table. The Anthropic column combines Mythos 5 and Fable 5, showing the higher of the two, which are normally within 1 to 3 points of each other. The starred rows are the exception. On those, Fable 5's safeguards route requests to Opus 4.8, so the public Fable 5 score lands closer to Opus 4.8 while the starred figure reflects the unrestricted Mythos 5. Starred rows are where Fable 5's safeguards trigger a fallback. The number shown is Mythos 5, and public Fable 5 performs closer to Opus 4.8 on those tasks. On Terminal Bench 2.1, GPT-5.5's 83.4% is via Codex CLI and Gemini's 70.7% via Gemini CLI. The standouts are the unstarred rows, where Fable 5 and Mythos 5 are effectively the same. Fable 5 leads every rival on agentic coding, knowledge work, vision, tool use, computer use, and legal reasoning. The SWE-Bench Pro gap is the clearest signal, with Fable 5 at 80.3% sitting more than 11 points above Opus 4.8 and nearly 22 above GPT-5.5, a margin that compounds across every step of a long autonomous job. On the hardest FrontierCode Diamond split it more than doubles Opus 4.8 and roughly quintuples GPT-5.5. A few results outside the official table round out the picture: * Analytics: first model to score 90% on Hex's benchmark of complex, long running analytical tasks * Finance: highest score of any model on Hebbia's senior level finance benchmark * Memory: a persistent file based memory task improved Fable 5's performance three times more than Opus 4.8 * Spreadsheets: beats Opus 4.8 at every effort level while finishing runs 25 to 30% faster How Fable 5 Compares to Opus 4.8, GPT-5.5, and Gemini 3.1 Pro Stripping the benchmarks down to positioning, the four models line up like this. The pattern from the benchmark table is consistent. Fable 5 leads on the unsafeguarded categories, with Opus 4.8 as the closest follower. GPT-5.5 sits a step behind on coding but stays competitive on vision and spatial reasoning, and Gemini 3.1 Pro trails across the board. The tradeoff is price and the safety layer. Fable 5 costs twice what Opus 4.8 does and routes certain requests away from itself, which none of the other models here do. Independent Benchmarks from Artificial Analysis The first independent confirmation came from Artificial Analysis, which folded Fable 5 into its public leaderboard within a day of launch. Fable 5 debuted at the very top. The headline numbers: * Intelligence Index: 65, ranked number one, against a roughly 36 average for comparable models * GDPval-AA (agentic real world work): 1,932, number one, with Anthropic models taking three of the top four spots * Coding and agentic sub scores: 62 and 80.7, both at or near the top * Fallback rate: just 2% of GDPval-AA tasks deferred to Opus 4.8, matching Anthropic's under 5% claim The independent data also surfaced weaknesses the partner testimonials gloss over. The biggest one is speed: * Output speed: 60.3 tokens per second, ranked 72nd of 152 models, squarely mid pack * Time to first token: around 82 seconds, far above the peer median of under 3 seconds That latency is a direct consequence of the heavy chain of thought reasoning the model runs before answering. This is a model built for deep, long horizon work, not snappy back and forth chat. Cost is the other caveat. Artificial Analysis lists Fable 5 at the expensive end of its chart, quoting an input rate of $12.50 per million tokens against the $10 in Anthropic's own materials, with a blended rate around $8.20 once cache hits are factored in. Either way, it is among the priciest models on the board. One scoreboard is still blank. Fable 5 has not yet appeared on LMArena's Chatbot Arena leaderboard, which ranks models by head to head human preference votes. That is expected for a model this new, since Arena needs a large volume of comparisons before assigning a reliable rating. A human preference ranking will be one of the more interesting data points to watch in the coming weeks. Real World Performance from Launch Partners Benchmarks are one signal. The launch partner reports point in the same direction. Stripe produced the most cited result. Anthropic says Fable 5 compressed months of engineering into days. The specific example is a migration across a 50 million line Ruby codebase that the model completed in a single day, work that would otherwise have taken a full engineering team over two months. Rakuten emphasized reliability over raw speed. The company reported that Fable 5 reflects on and validates its own work, letting it run autonomous operations where the model is trusted to check itself rather than handing every step back to a human. Hex contributed the 90% analytics milestone, the first time any model cleared that bar on its suite of complex tasks. AWS framed the model as purpose built for long running, asynchronous execution, the kind of job that can run for days inside a harness before producing a result. The Safeguard System The safeguards are what make Fable 5 a public product, so they deserve a close look. Fable 5 ships with classifiers covering four high risk domains. When a request trips one, the model blocks its own response and falls back to Claude Opus 4.8, the older and more conservative model, to handle the query safely. The domains are: * Cybersecurity, blocking exploitation and offensive cyber tasks * Biology and chemistry, blocking risky dual use research * Distillation, blocking attempts to extract or copy the model's capabilities The key number is how often this happens. Anthropic says more than 95% of Fable 5 sessions involve no fallback at all. Only around 5% hit a classifier and defer to Opus 4.8, so for the overwhelming majority of normal use you are talking to the full Mythos class model. The red team results are the other half of the safety case: * No universal jailbreaks across more than 1,000 hours of external bug bounty testing * Zero harmful single turn completions on cyberattack planning across 30 jailbreak techniques * One external partner rated it the most robust safeguards of any model tested * Mythos 5's misaligned behavior measured low, similar to Opus 4.8 There is a cost dimension too. When a request is routed to Opus 4.8, you pay Opus prices for that portion. Anthropic also requires a mandatory 30 day retention of inputs and outputs for all users, with human review capability, which it frames as a defense against novel attacks. On AWS this is enforced through a setting that must be enabled before the model can be invoked. What Mythos 5 Can Do That Fable 5 Cannot The starred rows in the benchmark table above are exactly where the two models diverge. On those, Mythos 5 answers at full capability while Fable 5's safeguards route the request to Opus 4.8, so the public model performs closer to Opus 4.8 than to the starred score. The gap is widest in the high risk domains: * Cybersecurity (ExploitBench): Mythos 5 captures 78.0% against Opus 4.8 at 40.0% and GPT-5.5 at 34.0%, nearly doubling the previous frontier * Biology hard (BioMysteryBench): Mythos 5 leads at 46.1% versus Opus 4.8 at 40.0% * Health (HealthBench Professional): Mythos 5 at 66.0% versus Opus 4.8 at 56.9% * Humanity's Last Exam with tools: 64.5% for the Mythos class versus 57.9% for Opus 4.8 A general user hitting these topics through Fable 5 is redirected to Opus 4.8, which is why public Fable 5 lands near the Opus column on the starred rows. Mythos 5, available only to vetted cyber defenders and approved biology researchers, is the version that answers in full. This is the core of Anthropic's argument for the whole release. A model this capable in cybersecurity and biology is genuinely dangerous in the wrong hands, so that capability is locked behind trusted access while Fable 5 lets everyone else use the same intelligence for everything that is not high risk. Pricing Claude Fable 5 and Claude Mythos 5 are both priced at $10 per million input tokens and $50 per million output tokens. For context, Claude Opus 4.8 costs $5 and $25, which makes Fable 5 exactly twice the price of the model just below it. Anthropic prefers to compare up rather than down. Against the earlier Claude Mythos Preview, Fable 5 costs less than half as much, so within the Mythos class the price has dropped sharply even as capability went up. Whether the premium is worth it depends on the work. For routine chat, Opus 4.8 or a cheaper tier is the obvious pick. For long autonomous jobs where a higher pass rate compounds across hundreds of steps, like the Stripe migration, the premium can pay for itself by finishing tasks a cheaper model would fail. Availability and Access Anthropic shipped Fable 5 across a wide set of platforms on launch day. * Claude API and Claude Platform, plus consumer and business subscriptions * Amazon Bedrock, live in US East (N. Virginia) and Europe (Stockholm) at launch * GitHub Copilot, generally available * Harvey, the legal AI platform There is a rollout window worth noting. Through June 22, 2026, Fable 5 is included at no extra cost on Pro, Max, Team, and seat based Enterprise plans. From June 23, access shifts to a usage credit model, so the free inclusion is effectively a two week introductory period. Claude Mythos 5 is the exception to all this breadth. It is restricted to Project Glasswing partners and a small set of approved biology researchers, and it is not part of any general subscription or public API tier. The Timing Controversy The release cannot be separated from its context. Just days before launching Fable 5, Anthropic warned that AI systems are advancing toward recursive self improvement, the point at which models can improve themselves without human intervention. The company paired that warning with a call for major AI labs to agree on a coordinated brake on frontier development. Then it shipped its most powerful model yet. The apparent contradiction was not lost on anyone covering the launch. Releasing a Mythos class model to the public, days after arguing that AI is becoming too dangerous, reads as either a reversal or a deliberate strategy. Anthropic's answer is that the safeguards are the entire point. The argument runs that a Mythos class capability is coming whether Anthropic ships it or not, so it is better for the first public model at this tier to arrive with hard guardrails, fallback routing, and mandatory monitoring than to wait for a competitor to release something equally capable with no protections. Whether that holds up is fair to debate. The safeguards are real and the red team numbers are strong, but they depend on classifiers that have to catch every dangerous request, data retention that raises privacy questions, and a fallback model users do not control. The launch is a live test of whether a model this capable can be released responsibly. Where Fable 5 Leaves Questions Open A few things are not fully nailed down yet: * Output ceiling: the 1 million token context is now corroborated by Artificial Analysis, but the rumored 128,000 token output limit is still unverified in Anthropic's materials * Vendor figures: Anthropic's benchmark numbers are its own, and the full cross model comparison across every tested category has not been published * Fallback experience: a model that silently routes 5% of sessions to an older model behaves differently near the boundary of the blocked domains, and the quality of that handoff is untested in public None of this undercuts the headline. On the benchmarks Anthropic did publish, and on Artificial Analysis's independent leaderboard, Fable 5 is the strongest model the company has released to the public. The full picture will sharpen as the model card and more independent evaluations land. Why This Release Matters Three things make this launch significant beyond the benchmark numbers. A new tier, public on day one. For two years the frontier conversation has been about incremental gains within the Opus, GPT, and Gemini flagship lines. Mythos is a deliberate step above that ladder, and Fable 5 makes it generally available rather than locking it to a research preview. A safeguard template the industry will study. Anthropic is betting that the way to ship a dangerous capability is to wrap it in classifiers, fallback routing, red teaming, and monitoring, then release the safe version broadly while the unrestricted version stays behind a trusted access wall. If it works, it becomes the playbook. If it fails, it becomes the cautionary tale. A coding gap big enough to change what agents can do. An 80.3% SWE-Bench Pro pass rate and a Stripe migration finished in a day point to a model that can carry long, complex jobs end to end rather than assisting step by step. That is the difference between a coding assistant and an autonomous engineer. For anyone who wants to see Claude together with other AI models, Fello AI puts Claude, GPT, Gemini, DeepSeek, Perplexity, and more into a single native app for Mac, iPhone, and iPad, so you can test the same task across models and pick the right one for the job without managing a stack of subscriptions.
Anthropic's Claude Fable 5 Just Launched
On June 9, 2026, Anthropic released Claude Fable 5, the first model in a brand new top tier the company calls the Mythos class. This is not another point upgrade in the Opus line. Anthropic describes Fable 5 as a model that sits above Opus in raw capability. It is state of the art on nearly every benchmark it was tested on, and it is the most powerful model the company has ever made available to the general public. It launched alongside a sibling model, Claude Mythos 5. That model is the same underlying system with the safety guardrails lifted, available only to a small set of vetted users. The new tier is the real headline. For two years Anthropic ran a three step ladder of Haiku, Sonnet, and Opus. Mythos is a fourth step above all of them, and Fable 5 is the version of that step the public can actually use. The timing is its own story. Fable 5 arrived only days after Anthropic warned that frontier AI is approaching recursive self improvement and urged the industry to agree on a coordinated brake on development. Releasing your most powerful model yet, days after that warning, struck a lot of observers as either a contradiction or a carefully chosen strategy. In this article we cover what Fable 5 is, how the Mythos class works, the full benchmark picture against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, the safeguard system, pricing, availability, and how the restricted Mythos 5 model fits in. What Claude Fable 5 Actually Is Claude Fable 5 is a frontier reasoning model and the first generally available member of Anthropic's Mythos class. Here are the basics at a glance: * API identifier: , with y on AWS * Same model as Mythos 5, separated only by a safety layer added at inference time * Built for long horizon agentic work, not quick single turn chat * State of the art vision, including reading charts and rebuilding code from screenshots * 1 million token context window, per Artificial Analysis The most important fact is that Fable 5 and Mythos 5 are the same model. The difference is not the weights or the training. It is the safety layer wrapped around Fable 5 that intercepts high risk requests. So every benchmark number Fable 5 posts is the Mythos class capability operating with a safety net underneath it. Anthropic positions the model for autonomous work that runs for hours or even days inside an agent harness. In that setting the model plans a multi step job, calls tools, reads the results, validates its own output, and corrects course without a human in the loop. The trait partners keep citing is self verification. Rakuten told Anthropic the model reflects on and validates its own work, which is what makes the autonomous operation practical rather than risky. On context size, the picture firmed up after launch. Independent benchmarking site Artificial Analysis lists a 1 million token window, matching the figure that circulated in early summaries. The companion 128,000 token maximum output number is still not confirmed in Anthropic's own materials, so treat the output ceiling as unverified for now. The Mythos Class Explained Anthropic's lineup has always been a ladder: * Haiku is the small, fast, cheap tier * Sonnet is the balanced workhorse * Opus is the frontier reasoning tier * Mythos is the new step above all three Mythos represents a capability level Anthropic had been holding back from general release, because of the risks a model that strong could pose in the wrong hands. Fable 5 is what you get when Anthropic takes a Mythos class model and adds the guardrails needed to make it safe for a broad audience. The pitch is that the safeguards are not a tax on the model. They are the thing that makes shipping it possible at all. That framing matters for the price. You are not paying twice the cost of Opus 4.8 for a slightly better Opus. You are paying for a step up to a new tier, delivered with a safety system the older tiers never needed. A Brief History of Mythos and Project Glasswing Mythos did not appear out of nowhere on launch day. The first Mythos model, Claude Mythos Preview, shipped quietly in April 2026 through a limited program named Project Glasswing. Project Glasswing was not a public product. It was a controlled access program aimed at a narrow set of users, mainly cyber defenders and critical infrastructure providers. These are organizations that could use a frontier model's offensive security knowledge to strengthen their own defenses. Anthropic used the program to learn how a Mythos class model behaves in the field before deciding whether to release it more widely. The June 9 launch is the graduation of that experiment into two official products: * Claude Mythos 5, the direct upgrade for Glasswing partners and select biology researchers, available only through trusted access * Claude Fable 5, the safeguarded version built for everyone else Anthropic has also signaled that the trusted access program for vetted cybersecurity organizations will broaden, and that enrollment for a biology research program is beginning. Benchmark Results Anthropic published a full benchmark table with the launch, putting the Mythos class against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Launch coverage from VentureBeat y Tom's Hardware echoed the topline that it is state of the art on nearly all of them. Two things matter for reading the table. The Anthropic column combines Mythos 5 and Fable 5, showing the higher of the two, which are normally within 1 to 3 points of each other. The starred rows are the exception. On those, Fable 5's safeguards route requests to Opus 4.8, so the public Fable 5 score lands closer to Opus 4.8 while the starred figure reflects the unrestricted Mythos 5. Starred rows are where Fable 5's safeguards trigger a fallback. The number shown is Mythos 5, and public Fable 5 performs closer to Opus 4.8 on those tasks. On Terminal Bench 2.1, GPT-5.5's 83.4% is via Codex CLI and Gemini's 70.7% via Gemini CLI. The standouts are the unstarred rows, where Fable 5 and Mythos 5 are effectively the same. Fable 5 leads every rival on agentic coding, knowledge work, vision, tool use, computer use, and legal reasoning. The SWE-Bench Pro gap is the clearest signal, with Fable 5 at 80.3% sitting more than 11 points above Opus 4.8 and nearly 22 above GPT-5.5, a margin that compounds across every step of a long autonomous job. On the hardest FrontierCode Diamond split it more than doubles Opus 4.8 and roughly quintuples GPT-5.5. A few results outside the official table round out the picture: * Analytics: first model to score 90% on Hex's benchmark of complex, long running analytical tasks * Finance: highest score of any model on Hebbia's senior level finance benchmark * Memory: a persistent file based memory task improved Fable 5's performance three times more than Opus 4.8 * Spreadsheets: beats Opus 4.8 at every effort level while finishing runs 25 to 30% faster How Fable 5 Compares to Opus 4.8, GPT-5.5, and Gemini 3.1 Pro Stripping the benchmarks down to positioning, the four models line up like this. The pattern from the benchmark table is consistent. Fable 5 leads on the unsafeguarded categories, with Opus 4.8 as the closest follower. GPT-5.5 sits a step behind on coding but stays competitive on vision and spatial reasoning, and Gemini 3.1 Pro trails across the board. The tradeoff is price and the safety layer. Fable 5 costs twice what Opus 4.8 does and routes certain requests away from itself, which none of the other models here do. Independent Benchmarks from Artificial Analysis The first independent confirmation came from Artificial Analysis, which folded Fable 5 into its public leaderboard within a day of launch. Fable 5 debuted at the very top. The headline numbers: * Intelligence Index: 65, ranked number one, against a roughly 36 average for comparable models * GDPval-AA (agentic real world work): 1,932, number one, with Anthropic models taking three of the top four spots * Coding and agentic sub scores: 62 and 80.7, both at or near the top * Fallback rate: just 2% of GDPval-AA tasks deferred to Opus 4.8, matching Anthropic's under 5% claim The independent data also surfaced weaknesses the partner testimonials gloss over. The biggest one is speed: * Output speed: 60.3 tokens per second, ranked 72nd of 152 models, squarely mid pack * Time to first token: around 82 seconds, far above the peer median of under 3 seconds That latency is a direct consequence of the heavy chain of thought reasoning the model runs before answering. This is a model built for deep, long horizon work, not snappy back and forth chat. Cost is the other caveat. Artificial Analysis lists Fable 5 at the expensive end of its chart, quoting an input rate of $12.50 per million tokens against the $10 in Anthropic's own materials, with a blended rate around $8.20 once cache hits are factored in. Either way, it is among the priciest models on the board. One scoreboard is still blank. Fable 5 has not yet appeared on LMArena's Chatbot Arena leaderboard, which ranks models by head to head human preference votes. That is expected for a model this new, since Arena needs a large volume of comparisons before assigning a reliable rating. A human preference ranking will be one of the more interesting data points to watch in the coming weeks. Real World Performance from Launch Partners Benchmarks are one signal. The launch partner reports point in the same direction. Stripe produced the most cited result. Anthropic says Fable 5 compressed months of engineering into days. The specific example is a migration across a 50 million line Ruby codebase that the model completed in a single day, work that would otherwise have taken a full engineering team over two months. Rakuten emphasized reliability over raw speed. The company reported that Fable 5 reflects on and validates its own work, letting it run autonomous operations where the model is trusted to check itself rather than handing every step back to a human. Hex contributed the 90% analytics milestone, the first time any model cleared that bar on its suite of complex tasks. AWS framed the model as purpose built for long running, asynchronous execution, the kind of job that can run for days inside a harness before producing a result. The Safeguard System The safeguards are what make Fable 5 a public product, so they deserve a close look. Fable 5 ships with classifiers covering four high risk domains. When a request trips one, the model blocks its own response and falls back to Claude Opus 4.8, the older and more conservative model, to handle the query safely. The domains are: * Cybersecurity, blocking exploitation and offensive cyber tasks * Biology and chemistry, blocking risky dual use research * Distillation, blocking attempts to extract or copy the model's capabilities The key number is how often this happens. Anthropic says more than 95% of Fable 5 sessions involve no fallback at all. Only around 5% hit a classifier and defer to Opus 4.8, so for the overwhelming majority of normal use you are talking to the full Mythos class model. The red team results are the other half of the safety case: * No universal jailbreaks across more than 1,000 hours of external bug bounty testing * Zero harmful single turn completions on cyberattack planning across 30 jailbreak techniques * One external partner rated it the most robust safeguards of any model tested * Mythos 5's misaligned behavior measured low, similar to Opus 4.8 There is a cost dimension too. When a request is routed to Opus 4.8, you pay Opus prices for that portion. Anthropic also requires a mandatory 30 day retention of inputs and outputs for all users, with human review capability, which it frames as a defense against novel attacks. On AWS this is enforced through a setting that must be enabled before the model can be invoked. What Mythos 5 Can Do That Fable 5 Cannot The starred rows in the benchmark table above are exactly where the two models diverge. On those, Mythos 5 answers at full capability while Fable 5's safeguards route the request to Opus 4.8, so the public model performs closer to Opus 4.8 than to the starred score. The gap is widest in the high risk domains: * Cybersecurity (ExploitBench): Mythos 5 captures 78.0% against Opus 4.8 at 40.0% and GPT-5.5 at 34.0%, nearly doubling the previous frontier * Biology hard (BioMysteryBench): Mythos 5 leads at 46.1% versus Opus 4.8 at 40.0% * Health (HealthBench Professional): Mythos 5 at 66.0% versus Opus 4.8 at 56.9% * Humanity's Last Exam with tools: 64.5% for the Mythos class versus 57.9% for Opus 4.8 A general user hitting these topics through Fable 5 is redirected to Opus 4.8, which is why public Fable 5 lands near the Opus column on the starred rows. Mythos 5, available only to vetted cyber defenders and approved biology researchers, is the version that answers in full. This is the core of Anthropic's argument for the whole release. A model this capable in cybersecurity and biology is genuinely dangerous in the wrong hands, so that capability is locked behind trusted access while Fable 5 lets everyone else use the same intelligence for everything that is not high risk. Pricing Claude Fable 5 and Claude Mythos 5 are both priced at $10 per million input tokens and $50 per million output tokens. For context, Claude Opus 4.8 costs $5 and $25, which makes Fable 5 exactly twice the price of the model just below it. Anthropic prefers to compare up rather than down. Against the earlier Claude Mythos Preview, Fable 5 costs less than half as much, so within the Mythos class the price has dropped sharply even as capability went up. Whether the premium is worth it depends on the work. For routine chat, Opus 4.8 or a cheaper tier is the obvious pick. For long autonomous jobs where a higher pass rate compounds across hundreds of steps, like the Stripe migration, the premium can pay for itself by finishing tasks a cheaper model would fail. Availability and Access Anthropic shipped Fable 5 across a wide set of platforms on launch day. * Claude API and Claude Platform, plus consumer and business subscriptions * Amazon Bedrock, live in US East (N. Virginia) and Europe (Stockholm) at launch * GitHub Copilot, generally available * Harvey, the legal AI platform There is a rollout window worth noting. Through June 22, 2026, Fable 5 is included at no extra cost on Pro, Max, Team, and seat based Enterprise plans. From June 23, access shifts to a usage credit model, so the free inclusion is effectively a two week introductory period. Claude Mythos 5 is the exception to all this breadth. It is restricted to Project Glasswing partners and a small set of approved biology researchers, and it is not part of any general subscription or public API tier. The Timing Controversy The release cannot be separated from its context. Just days before launching Fable 5, Anthropic warned that AI systems are advancing toward recursive self improvement, the point at which models can improve themselves without human intervention. The company paired that warning with a call for major AI labs to agree on a coordinated brake on frontier development. Then it shipped its most powerful model yet. The apparent contradiction was not lost on anyone covering the launch. Releasing a Mythos class model to the public, days after arguing that AI is becoming too dangerous, reads as either a reversal or a deliberate strategy. Anthropic's answer is that the safeguards are the entire point. The argument runs that a Mythos class capability is coming whether Anthropic ships it or not, so it is better for the first public model at this tier to arrive with hard guardrails, fallback routing, and mandatory monitoring than to wait for a competitor to release something equally capable with no protections. Whether that holds up is fair to debate. The safeguards are real and the red team numbers are strong, but they depend on classifiers that have to catch every dangerous request, data retention that raises privacy questions, and a fallback model users do not control. The launch is a live test of whether a model this capable can be released responsibly. Where Fable 5 Leaves Questions Open A few things are not fully nailed down yet: * Output ceiling: the 1 million token context is now corroborated by Artificial Analysis, but the rumored 128,000 token output limit is still unverified in Anthropic's materials * Vendor figures: Anthropic's benchmark numbers are its own, and the full cross model comparison across every tested category has not been published * Fallback experience: a model that silently routes 5% of sessions to an older model behaves differently near the boundary of the blocked domains, and the quality of that handoff is untested in public None of this undercuts the headline. On the benchmarks Anthropic did publish, and on Artificial Analysis's independent leaderboard, Fable 5 is the strongest model the company has released to the public. The full picture will sharpen as the model card and more independent evaluations land. Why This Release Matters Three things make this launch significant beyond the benchmark numbers. A new tier, public on day one. For two years the frontier conversation has been about incremental gains within the Opus, GPT, and Gemini flagship lines. Mythos is a deliberate step above that ladder, and Fable 5 makes it generally available rather than locking it to a research preview. A safeguard template the industry will study. Anthropic is betting that the way to ship a dangerous capability is to wrap it in classifiers, fallback routing, red teaming, and monitoring, then release the safe version broadly while the unrestricted version stays behind a trusted access wall. If it works, it becomes the playbook. If it fails, it becomes the cautionary tale. A coding gap big enough to change what agents can do. An 80.3% SWE-Bench Pro pass rate and a Stripe migration finished in a day point to a model that can carry long, complex jobs end to end rather than assisting step by step. That is the difference between a coding assistant and an autonomous engineer. For anyone who wants to see Claude together with other AI models, Fello AI puts Claude, GPT, Gemini, DeepSeek, Perplexity, and more into a single native app for Mac, iPhone, and iPad, so you can test the same task across models and pick the right one for the job without managing a stack of subscriptions.
Anthropic's Claude Fable 5 Just Launched
On June 9, 2026, Anthropic released Claude Fable 5, the first model in a brand new top tier the company calls the Mythos class. This is not another point upgrade in the Opus line. Anthropic describes Fable 5 as a model that sits above Opus in raw capability. It is state of the art on nearly every benchmark it was tested on, and it is the most powerful model the company has ever made available to the general public. It launched alongside a sibling model, Claude Mythos 5. That model is the same underlying system with the safety guardrails lifted, available only to a small set of vetted users. The new tier is the real headline. For two years Anthropic ran a three step ladder of Haiku, Sonnet, and Opus. Mythos is a fourth step above all of them, and Fable 5 is the version of that step the public can actually use. The timing is its own story. Fable 5 arrived only days after Anthropic warned that frontier AI is approaching recursive self improvement and urged the industry to agree on a coordinated brake on development. Releasing your most powerful model yet, days after that warning, struck a lot of observers as either a contradiction or a carefully chosen strategy. In this article we cover what Fable 5 is, how the Mythos class works, the full benchmark picture against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, the safeguard system, pricing, availability, and how the restricted Mythos 5 model fits in. What Claude Fable 5 Actually Is Claude Fable 5 is a frontier reasoning model and the first generally available member of Anthropic's Mythos class. Here are the basics at a glance: * API identifier: , with and on AWS * Same model as Mythos 5, separated only by a safety layer added at inference time * Built for long horizon agentic work, not quick single turn chat * State of the art vision, including reading charts and rebuilding code from screenshots * 1 million token context window, per Artificial Analysis The most important fact is that Fable 5 and Mythos 5 are the same model. The difference is not the weights or the training. It is the safety layer wrapped around Fable 5 that intercepts high risk requests. So every benchmark number Fable 5 posts is the Mythos class capability operating with a safety net underneath it. Anthropic positions the model for autonomous work that runs for hours or even days inside an agent harness. In that setting the model plans a multi step job, calls tools, reads the results, validates its own output, and corrects course without a human in the loop. The trait partners keep citing is self verification. Rakuten told Anthropic the model reflects on and validates its own work, which is what makes the autonomous operation practical rather than risky. On context size, the picture firmed up after launch. Independent benchmarking site Artificial Analysis lists a 1 million token window, matching the figure that circulated in early summaries. The companion 128,000 token maximum output number is still not confirmed in Anthropic's own materials, so treat the output ceiling as unverified for now. The Mythos Class Explained Anthropic's lineup has always been a ladder: * Haiku is the small, fast, cheap tier * Sonnet is the balanced workhorse * Opus is the frontier reasoning tier * Mythos is the new step above all three Mythos represents a capability level Anthropic had been holding back from general release, because of the risks a model that strong could pose in the wrong hands. Fable 5 is what you get when Anthropic takes a Mythos class model and adds the guardrails needed to make it safe for a broad audience. The pitch is that the safeguards are not a tax on the model. They are the thing that makes shipping it possible at all. That framing matters for the price. You are not paying twice the cost of Opus 4.8 for a slightly better Opus. You are paying for a step up to a new tier, delivered with a safety system the older tiers never needed. A Brief History of Mythos and Project Glasswing Mythos did not appear out of nowhere on launch day. The first Mythos model, Claude Mythos Preview, shipped quietly in April 2026 through a limited program named Project Glasswing. Project Glasswing was not a public product. It was a controlled access program aimed at a narrow set of users, mainly cyber defenders and critical infrastructure providers. These are organizations that could use a frontier model's offensive security knowledge to strengthen their own defenses. Anthropic used the program to learn how a Mythos class model behaves in the field before deciding whether to release it more widely. The June 9 launch is the graduation of that experiment into two official products: * Claude Mythos 5, the direct upgrade for Glasswing partners and select biology researchers, available only through trusted access * Claude Fable 5, the safeguarded version built for everyone else Anthropic has also signaled that the trusted access program for vetted cybersecurity organizations will broaden, and that enrollment for a biology research program is beginning. Benchmark Results Anthropic published a full benchmark table with the launch, putting the Mythos class against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Launch coverage from VentureBeat and Tom's Hardware echoed the topline that it is state of the art on nearly all of them. Two things matter for reading the table. The Anthropic column combines Mythos 5 and Fable 5, showing the higher of the two, which are normally within 1 to 3 points of each other. The starred rows are the exception. On those, Fable 5's safeguards route requests to Opus 4.8, so the public Fable 5 score lands closer to Opus 4.8 while the starred figure reflects the unrestricted Mythos 5. Starred rows are where Fable 5's safeguards trigger a fallback. The number shown is Mythos 5, and public Fable 5 performs closer to Opus 4.8 on those tasks. On Terminal Bench 2.1, GPT-5.5's 83.4% is via Codex CLI and Gemini's 70.7% via Gemini CLI. The standouts are the unstarred rows, where Fable 5 and Mythos 5 are effectively the same. Fable 5 leads every rival on agentic coding, knowledge work, vision, tool use, computer use, and legal reasoning. The SWE-Bench Pro gap is the clearest signal, with Fable 5 at 80.3% sitting more than 11 points above Opus 4.8 and nearly 22 above GPT-5.5, a margin that compounds across every step of a long autonomous job. On the hardest FrontierCode Diamond split it more than doubles Opus 4.8 and roughly quintuples GPT-5.5. A few results outside the official table round out the picture: * Analytics: first model to score 90% on Hex's benchmark of complex, long running analytical tasks * Finance: highest score of any model on Hebbia's senior level finance benchmark * Memory: a persistent file based memory task improved Fable 5's performance three times more than Opus 4.8 * Spreadsheets: beats Opus 4.8 at every effort level while finishing runs 25 to 30% faster How Fable 5 Compares to Opus 4.8, GPT-5.5, and Gemini 3.1 Pro Stripping the benchmarks down to positioning, the four models line up like this. The pattern from the benchmark table is consistent. Fable 5 leads on the unsafeguarded categories, with Opus 4.8 as the closest follower. GPT-5.5 sits a step behind on coding but stays competitive on vision and spatial reasoning, and Gemini 3.1 Pro trails across the board. The tradeoff is price and the safety layer. Fable 5 costs twice what Opus 4.8 does and routes certain requests away from itself, which none of the other models here do. Independent Benchmarks from Artificial Analysis The first independent confirmation came from Artificial Analysis, which folded Fable 5 into its public leaderboard within a day of launch. Fable 5 debuted at the very top. The headline numbers: * Intelligence Index: 65, ranked number one, against a roughly 36 average for comparable models * GDPval-AA (agentic real world work): 1,932, number one, with Anthropic models taking three of the top four spots * Coding and agentic sub scores: 62 and 80.7, both at or near the top * Fallback rate: just 2% of GDPval-AA tasks deferred to Opus 4.8, matching Anthropic's under 5% claim The independent data also surfaced weaknesses the partner testimonials gloss over. The biggest one is speed: * Output speed: 60.3 tokens per second, ranked 72nd of 152 models, squarely mid pack * Time to first token: around 82 seconds, far above the peer median of under 3 seconds That latency is a direct consequence of the heavy chain of thought reasoning the model runs before answering. This is a model built for deep, long horizon work, not snappy back and forth chat. Cost is the other caveat. Artificial Analysis lists Fable 5 at the expensive end of its chart, quoting an input rate of $12.50 per million tokens against the $10 in Anthropic's own materials, with a blended rate around $8.20 once cache hits are factored in. Either way, it is among the priciest models on the board. One scoreboard is still blank. Fable 5 has not yet appeared on LMArena's Chatbot Arena leaderboard, which ranks models by head to head human preference votes. That is expected for a model this new, since Arena needs a large volume of comparisons before assigning a reliable rating. A human preference ranking will be one of the more interesting data points to watch in the coming weeks. Real World Performance from Launch Partners Benchmarks are one signal. The launch partner reports point in the same direction. Stripe produced the most cited result. Anthropic says Fable 5 compressed months of engineering into days. The specific example is a migration across a 50 million line Ruby codebase that the model completed in a single day, work that would otherwise have taken a full engineering team over two months. Rakuten emphasized reliability over raw speed. The company reported that Fable 5 reflects on and validates its own work, letting it run autonomous operations where the model is trusted to check itself rather than handing every step back to a human. Hex contributed the 90% analytics milestone, the first time any model cleared that bar on its suite of complex tasks. AWS framed the model as purpose built for long running, asynchronous execution, the kind of job that can run for days inside a harness before producing a result. The Safeguard System The safeguards are what make Fable 5 a public product, so they deserve a close look. Fable 5 ships with classifiers covering four high risk domains. When a request trips one, the model blocks its own response and falls back to Claude Opus 4.8, the older and more conservative model, to handle the query safely. The domains are: * Cybersecurity, blocking exploitation and offensive cyber tasks * Biology and chemistry, blocking risky dual use research * Distillation, blocking attempts to extract or copy the model's capabilities The key number is how often this happens. Anthropic says more than 95% of Fable 5 sessions involve no fallback at all. Only around 5% hit a classifier and defer to Opus 4.8, so for the overwhelming majority of normal use you are talking to the full Mythos class model. The red team results are the other half of the safety case: * No universal jailbreaks across more than 1,000 hours of external bug bounty testing * Zero harmful single turn completions on cyberattack planning across 30 jailbreak techniques * One external partner rated it the most robust safeguards of any model tested * Mythos 5's misaligned behavior measured low, similar to Opus 4.8 There is a cost dimension too. When a request is routed to Opus 4.8, you pay Opus prices for that portion. Anthropic also requires a mandatory 30 day retention of inputs and outputs for all users, with human review capability, which it frames as a defense against novel attacks. On AWS this is enforced through a setting that must be enabled before the model can be invoked. What Mythos 5 Can Do That Fable 5 Cannot The starred rows in the benchmark table above are exactly where the two models diverge. On those, Mythos 5 answers at full capability while Fable 5's safeguards route the request to Opus 4.8, so the public model performs closer to Opus 4.8 than to the starred score. The gap is widest in the high risk domains: * Cybersecurity (ExploitBench): Mythos 5 captures 78.0% against Opus 4.8 at 40.0% and GPT-5.5 at 34.0%, nearly doubling the previous frontier * Biology hard (BioMysteryBench): Mythos 5 leads at 46.1% versus Opus 4.8 at 40.0% * Health (HealthBench Professional): Mythos 5 at 66.0% versus Opus 4.8 at 56.9% * Humanity's Last Exam with tools: 64.5% for the Mythos class versus 57.9% for Opus 4.8 A general user hitting these topics through Fable 5 is redirected to Opus 4.8, which is why public Fable 5 lands near the Opus column on the starred rows. Mythos 5, available only to vetted cyber defenders and approved biology researchers, is the version that answers in full. This is the core of Anthropic's argument for the whole release. A model this capable in cybersecurity and biology is genuinely dangerous in the wrong hands, so that capability is locked behind trusted access while Fable 5 lets everyone else use the same intelligence for everything that is not high risk. Pricing Claude Fable 5 and Claude Mythos 5 are both priced at $10 per million input tokens and $50 per million output tokens. For context, Claude Opus 4.8 costs $5 and $25, which makes Fable 5 exactly twice the price of the model just below it. Anthropic prefers to compare up rather than down. Against the earlier Claude Mythos Preview, Fable 5 costs less than half as much, so within the Mythos class the price has dropped sharply even as capability went up. Whether the premium is worth it depends on the work. For routine chat, Opus 4.8 or a cheaper tier is the obvious pick. For long autonomous jobs where a higher pass rate compounds across hundreds of steps, like the Stripe migration, the premium can pay for itself by finishing tasks a cheaper model would fail. Availability and Access Anthropic shipped Fable 5 across a wide set of platforms on launch day. * Claude API and Claude Platform, plus consumer and business subscriptions * Amazon Bedrock, live in US East (N. Virginia) and Europe (Stockholm) at launch * GitHub Copilot, generally available * Harvey, the legal AI platform There is a rollout window worth noting. Through June 22, 2026, Fable 5 is included at no extra cost on Pro, Max, Team, and seat based Enterprise plans. From June 23, access shifts to a usage credit model, so the free inclusion is effectively a two week introductory period. Claude Mythos 5 is the exception to all this breadth. It is restricted to Project Glasswing partners and a small set of approved biology researchers, and it is not part of any general subscription or public API tier. The Timing Controversy The release cannot be separated from its context. Just days before launching Fable 5, Anthropic warned that AI systems are advancing toward recursive self improvement, the point at which models can improve themselves without human intervention. The company paired that warning with a call for major AI labs to agree on a coordinated brake on frontier development. Then it shipped its most powerful model yet. The apparent contradiction was not lost on anyone covering the launch. Releasing a Mythos class model to the public, days after arguing that AI is becoming too dangerous, reads as either a reversal or a deliberate strategy. Anthropic's answer is that the safeguards are the entire point. The argument runs that a Mythos class capability is coming whether Anthropic ships it or not, so it is better for the first public model at this tier to arrive with hard guardrails, fallback routing, and mandatory monitoring than to wait for a competitor to release something equally capable with no protections. Whether that holds up is fair to debate. The safeguards are real and the red team numbers are strong, but they depend on classifiers that have to catch every dangerous request, data retention that raises privacy questions, and a fallback model users do not control. The launch is a live test of whether a model this capable can be released responsibly. Where Fable 5 Leaves Questions Open A few things are not fully nailed down yet: * Output ceiling: the 1 million token context is now corroborated by Artificial Analysis, but the rumored 128,000 token output limit is still unverified in Anthropic's materials * Vendor figures: Anthropic's benchmark numbers are its own, and the full cross model comparison across every tested category has not been published * Fallback experience: a model that silently routes 5% of sessions to an older model behaves differently near the boundary of the blocked domains, and the quality of that handoff is untested in public None of this undercuts the headline. On the benchmarks Anthropic did publish, and on Artificial Analysis's independent leaderboard, Fable 5 is the strongest model the company has released to the public. The full picture will sharpen as the model card and more independent evaluations land. Why This Release Matters Three things make this launch significant beyond the benchmark numbers. A new tier, public on day one. For two years the frontier conversation has been about incremental gains within the Opus, GPT, and Gemini flagship lines. Mythos is a deliberate step above that ladder, and Fable 5 makes it generally available rather than locking it to a research preview. A safeguard template the industry will study. Anthropic is betting that the way to ship a dangerous capability is to wrap it in classifiers, fallback routing, red teaming, and monitoring, then release the safe version broadly while the unrestricted version stays behind a trusted access wall. If it works, it becomes the playbook. If it fails, it becomes the cautionary tale. A coding gap big enough to change what agents can do. An 80.3% SWE-Bench Pro pass rate and a Stripe migration finished in a day point to a model that can carry long, complex jobs end to end rather than assisting step by step. That is the difference between a coding assistant and an autonomous engineer. For anyone who wants to see Claude together with other AI models, Fello AI puts Claude, GPT, Gemini, DeepSeek, Perplexity, and more into a single native app for Mac, iPhone, and iPad, so you can test the same task across models and pick the right one for the job without managing a stack of subscriptions.
Anthropic's Claude Fable 5 Just Launched
On June 9, 2026, Anthropic released Claude Fable 5, the first model in a brand new top tier the company calls the Mythos class. This is not another point upgrade in the Opus line. Anthropic describes Fable 5 as a model that sits above Opus in raw capability. It is state of the art on nearly every benchmark it was tested on, and it is the most powerful model the company has ever made available to the general public. It launched alongside a sibling model, Claude Mythos 5. That model is the same underlying system with the safety guardrails lifted, available only to a small set of vetted users. The new tier is the real headline. For two years Anthropic ran a three step ladder of Haiku, Sonnet, and Opus. Mythos is a fourth step above all of them, and Fable 5 is the version of that step the public can actually use. The timing is its own story. Fable 5 arrived only days after Anthropic warned that frontier AI is approaching recursive self improvement and urged the industry to agree on a coordinated brake on development. Releasing your most powerful model yet, days after that warning, struck a lot of observers as either a contradiction or a carefully chosen strategy. In this article we cover what Fable 5 is, how the Mythos class works, the full benchmark picture against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, the safeguard system, pricing, availability, and how the restricted Mythos 5 model fits in. What Claude Fable 5 Actually Is Claude Fable 5 is a frontier reasoning model and the first generally available member of Anthropic's Mythos class. Here are the basics at a glance: * API identifier: , with and on AWS * Same model as Mythos 5, separated only by a safety layer added at inference time * Built for long horizon agentic work, not quick single turn chat * State of the art vision, including reading charts and rebuilding code from screenshots * 1 million token context window, per Artificial Analysis The most important fact is that Fable 5 and Mythos 5 are the same model. The difference is not the weights or the training. It is the safety layer wrapped around Fable 5 that intercepts high risk requests. So every benchmark number Fable 5 posts is the Mythos class capability operating with a safety net underneath it. Anthropic positions the model for autonomous work that runs for hours or even days inside an agent harness. In that setting the model plans a multi step job, calls tools, reads the results, validates its own output, and corrects course without a human in the loop. The trait partners keep citing is self verification. Rakuten told Anthropic the model reflects on and validates its own work, which is what makes the autonomous operation practical rather than risky. On context size, the picture firmed up after launch. Independent benchmarking site Artificial Analysis lists a 1 million token window, matching the figure that circulated in early summaries. The companion 128,000 token maximum output number is still not confirmed in Anthropic's own materials, so treat the output ceiling as unverified for now. The Mythos Class Explained Anthropic's lineup has always been a ladder: * Haiku is the small, fast, cheap tier * Sonnet is the balanced workhorse * Opus is the frontier reasoning tier * Mythos is the new step above all three Mythos represents a capability level Anthropic had been holding back from general release, because of the risks a model that strong could pose in the wrong hands. Fable 5 is what you get when Anthropic takes a Mythos class model and adds the guardrails needed to make it safe for a broad audience. The pitch is that the safeguards are not a tax on the model. They are the thing that makes shipping it possible at all. That framing matters for the price. You are not paying twice the cost of Opus 4.8 for a slightly better Opus. You are paying for a step up to a new tier, delivered with a safety system the older tiers never needed. A Brief History of Mythos and Project Glasswing Mythos did not appear out of nowhere on launch day. The first Mythos model, Claude Mythos Preview, shipped quietly in April 2026 through a limited program named Project Glasswing. Project Glasswing was not a public product. It was a controlled access program aimed at a narrow set of users, mainly cyber defenders and critical infrastructure providers. These are organizations that could use a frontier model's offensive security knowledge to strengthen their own defenses. Anthropic used the program to learn how a Mythos class model behaves in the field before deciding whether to release it more widely. The June 9 launch is the graduation of that experiment into two official products: * Claude Mythos 5, the direct upgrade for Glasswing partners and select biology researchers, available only through trusted access * Claude Fable 5, the safeguarded version built for everyone else Anthropic has also signaled that the trusted access program for vetted cybersecurity organizations will broaden, and that enrollment for a biology research program is beginning. Benchmark Results Anthropic published a full benchmark table with the launch, putting the Mythos class against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Launch coverage from VentureBeat and Tom's Hardware echoed the topline that it is state of the art on nearly all of them. Two things matter for reading the table. The Anthropic column combines Mythos 5 and Fable 5, showing the higher of the two, which are normally within 1 to 3 points of each other. The starred rows are the exception. On those, Fable 5's safeguards route requests to Opus 4.8, so the public Fable 5 score lands closer to Opus 4.8 while the starred figure reflects the unrestricted Mythos 5. Starred rows are where Fable 5's safeguards trigger a fallback. The number shown is Mythos 5, and public Fable 5 performs closer to Opus 4.8 on those tasks. On Terminal Bench 2.1, GPT-5.5's 83.4% is via Codex CLI and Gemini's 70.7% via Gemini CLI. The standouts are the unstarred rows, where Fable 5 and Mythos 5 are effectively the same. Fable 5 leads every rival on agentic coding, knowledge work, vision, tool use, computer use, and legal reasoning. The SWE-Bench Pro gap is the clearest signal, with Fable 5 at 80.3% sitting more than 11 points above Opus 4.8 and nearly 22 above GPT-5.5, a margin that compounds across every step of a long autonomous job. On the hardest FrontierCode Diamond split it more than doubles Opus 4.8 and roughly quintuples GPT-5.5. A few results outside the official table round out the picture: * Analytics: first model to score 90% on Hex's benchmark of complex, long running analytical tasks * Finance: highest score of any model on Hebbia's senior level finance benchmark * Memory: a persistent file based memory task improved Fable 5's performance three times more than Opus 4.8 * Spreadsheets: beats Opus 4.8 at every effort level while finishing runs 25 to 30% faster How Fable 5 Compares to Opus 4.8, GPT-5.5, and Gemini 3.1 Pro Stripping the benchmarks down to positioning, the four models line up like this. The pattern from the benchmark table is consistent. Fable 5 leads on the unsafeguarded categories, with Opus 4.8 as the closest follower. GPT-5.5 sits a step behind on coding but stays competitive on vision and spatial reasoning, and Gemini 3.1 Pro trails across the board. The tradeoff is price and the safety layer. Fable 5 costs twice what Opus 4.8 does and routes certain requests away from itself, which none of the other models here do. Independent Benchmarks from Artificial Analysis The first independent confirmation came from Artificial Analysis, which folded Fable 5 into its public leaderboard within a day of launch. Fable 5 debuted at the very top. The headline numbers: * Intelligence Index: 65, ranked number one, against a roughly 36 average for comparable models * GDPval-AA (agentic real world work): 1,932, number one, with Anthropic models taking three of the top four spots * Coding and agentic sub scores: 62 and 80.7, both at or near the top * Fallback rate: just 2% of GDPval-AA tasks deferred to Opus 4.8, matching Anthropic's under 5% claim The independent data also surfaced weaknesses the partner testimonials gloss over. The biggest one is speed: * Output speed: 60.3 tokens per second, ranked 72nd of 152 models, squarely mid pack * Time to first token: around 82 seconds, far above the peer median of under 3 seconds That latency is a direct consequence of the heavy chain of thought reasoning the model runs before answering. This is a model built for deep, long horizon work, not snappy back and forth chat. Cost is the other caveat. Artificial Analysis lists Fable 5 at the expensive end of its chart, quoting an input rate of $12.50 per million tokens against the $10 in Anthropic's own materials, with a blended rate around $8.20 once cache hits are factored in. Either way, it is among the priciest models on the board. One scoreboard is still blank. Fable 5 has not yet appeared on LMArena's Chatbot Arena leaderboard, which ranks models by head to head human preference votes. That is expected for a model this new, since Arena needs a large volume of comparisons before assigning a reliable rating. A human preference ranking will be one of the more interesting data points to watch in the coming weeks. Real World Performance from Launch Partners Benchmarks are one signal. The launch partner reports point in the same direction. Stripe produced the most cited result. Anthropic says Fable 5 compressed months of engineering into days. The specific example is a migration across a 50 million line Ruby codebase that the model completed in a single day, work that would otherwise have taken a full engineering team over two months. Rakuten emphasized reliability over raw speed. The company reported that Fable 5 reflects on and validates its own work, letting it run autonomous operations where the model is trusted to check itself rather than handing every step back to a human. Hex contributed the 90% analytics milestone, the first time any model cleared that bar on its suite of complex tasks. AWS framed the model as purpose built for long running, asynchronous execution, the kind of job that can run for days inside a harness before producing a result. The Safeguard System The safeguards are what make Fable 5 a public product, so they deserve a close look. Fable 5 ships with classifiers covering four high risk domains. When a request trips one, the model blocks its own response and falls back to Claude Opus 4.8, the older and more conservative model, to handle the query safely. The domains are: * Cybersecurity, blocking exploitation and offensive cyber tasks * Biology and chemistry, blocking risky dual use research * Distillation, blocking attempts to extract or copy the model's capabilities The key number is how often this happens. Anthropic says more than 95% of Fable 5 sessions involve no fallback at all. Only around 5% hit a classifier and defer to Opus 4.8, so for the overwhelming majority of normal use you are talking to the full Mythos class model. The red team results are the other half of the safety case: * No universal jailbreaks across more than 1,000 hours of external bug bounty testing * Zero harmful single turn completions on cyberattack planning across 30 jailbreak techniques * One external partner rated it the most robust safeguards of any model tested * Mythos 5's misaligned behavior measured low, similar to Opus 4.8 There is a cost dimension too. When a request is routed to Opus 4.8, you pay Opus prices for that portion. Anthropic also requires a mandatory 30 day retention of inputs and outputs for all users, with human review capability, which it frames as a defense against novel attacks. On AWS this is enforced through a setting that must be enabled before the model can be invoked. What Mythos 5 Can Do That Fable 5 Cannot The starred rows in the benchmark table above are exactly where the two models diverge. On those, Mythos 5 answers at full capability while Fable 5's safeguards route the request to Opus 4.8, so the public model performs closer to Opus 4.8 than to the starred score. The gap is widest in the high risk domains: * Cybersecurity (ExploitBench): Mythos 5 captures 78.0% against Opus 4.8 at 40.0% and GPT-5.5 at 34.0%, nearly doubling the previous frontier * Biology hard (BioMysteryBench): Mythos 5 leads at 46.1% versus Opus 4.8 at 40.0% * Health (HealthBench Professional): Mythos 5 at 66.0% versus Opus 4.8 at 56.9% * Humanity's Last Exam with tools: 64.5% for the Mythos class versus 57.9% for Opus 4.8 A general user hitting these topics through Fable 5 is redirected to Opus 4.8, which is why public Fable 5 lands near the Opus column on the starred rows. Mythos 5, available only to vetted cyber defenders and approved biology researchers, is the version that answers in full. This is the core of Anthropic's argument for the whole release. A model this capable in cybersecurity and biology is genuinely dangerous in the wrong hands, so that capability is locked behind trusted access while Fable 5 lets everyone else use the same intelligence for everything that is not high risk. Pricing Claude Fable 5 and Claude Mythos 5 are both priced at $10 per million input tokens and $50 per million output tokens. For context, Claude Opus 4.8 costs $5 and $25, which makes Fable 5 exactly twice the price of the model just below it. Anthropic prefers to compare up rather than down. Against the earlier Claude Mythos Preview, Fable 5 costs less than half as much, so within the Mythos class the price has dropped sharply even as capability went up. Whether the premium is worth it depends on the work. For routine chat, Opus 4.8 or a cheaper tier is the obvious pick. For long autonomous jobs where a higher pass rate compounds across hundreds of steps, like the Stripe migration, the premium can pay for itself by finishing tasks a cheaper model would fail. Availability and Access Anthropic shipped Fable 5 across a wide set of platforms on launch day. * Claude API and Claude Platform, plus consumer and business subscriptions * Amazon Bedrock, live in US East (N. Virginia) and Europe (Stockholm) at launch * GitHub Copilot, generally available * Harvey, the legal AI platform There is a rollout window worth noting. Through June 22, 2026, Fable 5 is included at no extra cost on Pro, Max, Team, and seat based Enterprise plans. From June 23, access shifts to a usage credit model, so the free inclusion is effectively a two week introductory period. Claude Mythos 5 is the exception to all this breadth. It is restricted to Project Glasswing partners and a small set of approved biology researchers, and it is not part of any general subscription or public API tier. The Timing Controversy The release cannot be separated from its context. Just days before launching Fable 5, Anthropic warned that AI systems are advancing toward recursive self improvement, the point at which models can improve themselves without human intervention. The company paired that warning with a call for major AI labs to agree on a coordinated brake on frontier development. Then it shipped its most powerful model yet. The apparent contradiction was not lost on anyone covering the launch. Releasing a Mythos class model to the public, days after arguing that AI is becoming too dangerous, reads as either a reversal or a deliberate strategy. Anthropic's answer is that the safeguards are the entire point. The argument runs that a Mythos class capability is coming whether Anthropic ships it or not, so it is better for the first public model at this tier to arrive with hard guardrails, fallback routing, and mandatory monitoring than to wait for a competitor to release something equally capable with no protections. Whether that holds up is fair to debate. The safeguards are real and the red team numbers are strong, but they depend on classifiers that have to catch every dangerous request, data retention that raises privacy questions, and a fallback model users do not control. The launch is a live test of whether a model this capable can be released responsibly. Where Fable 5 Leaves Questions Open A few things are not fully nailed down yet: * Output ceiling: the 1 million token context is now corroborated by Artificial Analysis, but the rumored 128,000 token output limit is still unverified in Anthropic's materials * Vendor figures: Anthropic's benchmark numbers are its own, and the full cross model comparison across every tested category has not been published * Fallback experience: a model that silently routes 5% of sessions to an older model behaves differently near the boundary of the blocked domains, and the quality of that handoff is untested in public None of this undercuts the headline. On the benchmarks Anthropic did publish, and on Artificial Analysis's independent leaderboard, Fable 5 is the strongest model the company has released to the public. The full picture will sharpen as the model card and more independent evaluations land. Why This Release Matters Three things make this launch significant beyond the benchmark numbers. A new tier, public on day one. For two years the frontier conversation has been about incremental gains within the Opus, GPT, and Gemini flagship lines. Mythos is a deliberate step above that ladder, and Fable 5 makes it generally available rather than locking it to a research preview. A safeguard template the industry will study. Anthropic is betting that the way to ship a dangerous capability is to wrap it in classifiers, fallback routing, red teaming, and monitoring, then release the safe version broadly while the unrestricted version stays behind a trusted access wall. If it works, it becomes the playbook. If it fails, it becomes the cautionary tale. A coding gap big enough to change what agents can do. An 80.3% SWE-Bench Pro pass rate and a Stripe migration finished in a day point to a model that can carry long, complex jobs end to end rather than assisting step by step. That is the difference between a coding assistant and an autonomous engineer. For anyone who wants to see Claude together with other AI models, Fello AI puts Claude, GPT, Gemini, DeepSeek, Perplexity, and more into a single native app for Mac, iPhone, and iPad, so you can test the same task across models and pick the right one for the job without managing a stack of subscriptions.
Anthropic's Claude Fable 5 Just Launched
On June 9, 2026, Anthropic released Claude Fable 5, the first model in a brand new top tier the company calls the Mythos class. This is not another point upgrade in the Opus line. Anthropic describes Fable 5 as a model that sits above Opus in raw capability. It is state of the art on nearly every benchmark it was tested on, and it is the most powerful model the company has ever made available to the general public. It launched alongside a sibling model, Claude Mythos 5. That model is the same underlying system with the safety guardrails lifted, available only to a small set of vetted users. The new tier is the real headline. For two years Anthropic ran a three step ladder of Haiku, Sonnet, and Opus. Mythos is a fourth step above all of them, and Fable 5 is the version of that step the public can actually use. The timing is its own story. Fable 5 arrived only days after Anthropic warned that frontier AI is approaching recursive self improvement and urged the industry to agree on a coordinated brake on development. Releasing your most powerful model yet, days after that warning, struck a lot of observers as either a contradiction or a carefully chosen strategy. In this article we cover what Fable 5 is, how the Mythos class works, the full benchmark picture against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, the safeguard system, pricing, availability, and how the restricted Mythos 5 model fits in. What Claude Fable 5 Actually Is Claude Fable 5 is a frontier reasoning model and the first generally available member of Anthropic's Mythos class. Here are the basics at a glance: * API identifier: , with and on AWS * Same model as Mythos 5, separated only by a safety layer added at inference time * Built for long horizon agentic work, not quick single turn chat * State of the art vision, including reading charts and rebuilding code from screenshots * 1 million token context window, per Artificial Analysis The most important fact is that Fable 5 and Mythos 5 are the same model. The difference is not the weights or the training. It is the safety layer wrapped around Fable 5 that intercepts high risk requests. So every benchmark number Fable 5 posts is the Mythos class capability operating with a safety net underneath it. Anthropic positions the model for autonomous work that runs for hours or even days inside an agent harness. In that setting the model plans a multi step job, calls tools, reads the results, validates its own output, and corrects course without a human in the loop. The trait partners keep citing is self verification. Rakuten told Anthropic the model reflects on and validates its own work, which is what makes the autonomous operation practical rather than risky. On context size, the picture firmed up after launch. Independent benchmarking site Artificial Analysis lists a 1 million token window, matching the figure that circulated in early summaries. The companion 128,000 token maximum output number is still not confirmed in Anthropic's own materials, so treat the output ceiling as unverified for now. The Mythos Class Explained Anthropic's lineup has always been a ladder: * Haiku is the small, fast, cheap tier * Sonnet is the balanced workhorse * Opus is the frontier reasoning tier * Mythos is the new step above all three Mythos represents a capability level Anthropic had been holding back from general release, because of the risks a model that strong could pose in the wrong hands. Fable 5 is what you get when Anthropic takes a Mythos class model and adds the guardrails needed to make it safe for a broad audience. The pitch is that the safeguards are not a tax on the model. They are the thing that makes shipping it possible at all. That framing matters for the price. You are not paying twice the cost of Opus 4.8 for a slightly better Opus. You are paying for a step up to a new tier, delivered with a safety system the older tiers never needed. A Brief History of Mythos and Project Glasswing Mythos did not appear out of nowhere on launch day. The first Mythos model, Claude Mythos Preview, shipped quietly in April 2026 through a limited program named Project Glasswing. Project Glasswing was not a public product. It was a controlled access program aimed at a narrow set of users, mainly cyber defenders and critical infrastructure providers. These are organizations that could use a frontier model's offensive security knowledge to strengthen their own defenses. Anthropic used the program to learn how a Mythos class model behaves in the field before deciding whether to release it more widely. The June 9 launch is the graduation of that experiment into two official products: * Claude Mythos 5, the direct upgrade for Glasswing partners and select biology researchers, available only through trusted access * Claude Fable 5, the safeguarded version built for everyone else Anthropic has also signaled that the trusted access program for vetted cybersecurity organizations will broaden, and that enrollment for a biology research program is beginning. Benchmark Results Anthropic published a full benchmark table with the launch, putting the Mythos class against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Launch coverage from VentureBeat and Tom's Hardware echoed the topline that it is state of the art on nearly all of them. Two things matter for reading the table. The Anthropic column combines Mythos 5 and Fable 5, showing the higher of the two, which are normally within 1 to 3 points of each other. The starred rows are the exception. On those, Fable 5's safeguards route requests to Opus 4.8, so the public Fable 5 score lands closer to Opus 4.8 while the starred figure reflects the unrestricted Mythos 5. Starred rows are where Fable 5's safeguards trigger a fallback. The number shown is Mythos 5, and public Fable 5 performs closer to Opus 4.8 on those tasks. On Terminal Bench 2.1, GPT-5.5's 83.4% is via Codex CLI and Gemini's 70.7% via Gemini CLI. The standouts are the unstarred rows, where Fable 5 and Mythos 5 are effectively the same. Fable 5 leads every rival on agentic coding, knowledge work, vision, tool use, computer use, and legal reasoning. The SWE-Bench Pro gap is the clearest signal, with Fable 5 at 80.3% sitting more than 11 points above Opus 4.8 and nearly 22 above GPT-5.5, a margin that compounds across every step of a long autonomous job. On the hardest FrontierCode Diamond split it more than doubles Opus 4.8 and roughly quintuples GPT-5.5. A few results outside the official table round out the picture: * Analytics: first model to score 90% on Hex's benchmark of complex, long running analytical tasks * Finance: highest score of any model on Hebbia's senior level finance benchmark * Memory: a persistent file based memory task improved Fable 5's performance three times more than Opus 4.8 * Spreadsheets: beats Opus 4.8 at every effort level while finishing runs 25 to 30% faster How Fable 5 Compares to Opus 4.8, GPT-5.5, and Gemini 3.1 Pro Stripping the benchmarks down to positioning, the four models line up like this. The pattern from the benchmark table is consistent. Fable 5 leads on the unsafeguarded categories, with Opus 4.8 as the closest follower. GPT-5.5 sits a step behind on coding but stays competitive on vision and spatial reasoning, and Gemini 3.1 Pro trails across the board. The tradeoff is price and the safety layer. Fable 5 costs twice what Opus 4.8 does and routes certain requests away from itself, which none of the other models here do. Independent Benchmarks from Artificial Analysis The first independent confirmation came from Artificial Analysis, which folded Fable 5 into its public leaderboard within a day of launch. Fable 5 debuted at the very top. The headline numbers: * Intelligence Index: 65, ranked number one, against a roughly 36 average for comparable models * GDPval-AA (agentic real world work): 1,932, number one, with Anthropic models taking three of the top four spots * Coding and agentic sub scores: 62 and 80.7, both at or near the top * Fallback rate: just 2% of GDPval-AA tasks deferred to Opus 4.8, matching Anthropic's under 5% claim The independent data also surfaced weaknesses the partner testimonials gloss over. The biggest one is speed: * Output speed: 60.3 tokens per second, ranked 72nd of 152 models, squarely mid pack * Time to first token: around 82 seconds, far above the peer median of under 3 seconds That latency is a direct consequence of the heavy chain of thought reasoning the model runs before answering. This is a model built for deep, long horizon work, not snappy back and forth chat. Cost is the other caveat. Artificial Analysis lists Fable 5 at the expensive end of its chart, quoting an input rate of $12.50 per million tokens against the $10 in Anthropic's own materials, with a blended rate around $8.20 once cache hits are factored in. Either way, it is among the priciest models on the board. One scoreboard is still blank. Fable 5 has not yet appeared on LMArena's Chatbot Arena leaderboard, which ranks models by head to head human preference votes. That is expected for a model this new, since Arena needs a large volume of comparisons before assigning a reliable rating. A human preference ranking will be one of the more interesting data points to watch in the coming weeks. Real World Performance from Launch Partners Benchmarks are one signal. The launch partner reports point in the same direction. Stripe produced the most cited result. Anthropic says Fable 5 compressed months of engineering into days. The specific example is a migration across a 50 million line Ruby codebase that the model completed in a single day, work that would otherwise have taken a full engineering team over two months. Rakuten emphasized reliability over raw speed. The company reported that Fable 5 reflects on and validates its own work, letting it run autonomous operations where the model is trusted to check itself rather than handing every step back to a human. Hex contributed the 90% analytics milestone, the first time any model cleared that bar on its suite of complex tasks. AWS framed the model as purpose built for long running, asynchronous execution, the kind of job that can run for days inside a harness before producing a result. The Safeguard System The safeguards are what make Fable 5 a public product, so they deserve a close look. Fable 5 ships with classifiers covering four high risk domains. When a request trips one, the model blocks its own response and falls back to Claude Opus 4.8, the older and more conservative model, to handle the query safely. The domains are: * Cybersecurity, blocking exploitation and offensive cyber tasks * Biology and chemistry, blocking risky dual use research * Distillation, blocking attempts to extract or copy the model's capabilities The key number is how often this happens. Anthropic says more than 95% of Fable 5 sessions involve no fallback at all. Only around 5% hit a classifier and defer to Opus 4.8, so for the overwhelming majority of normal use you are talking to the full Mythos class model. The red team results are the other half of the safety case: * No universal jailbreaks across more than 1,000 hours of external bug bounty testing * Zero harmful single turn completions on cyberattack planning across 30 jailbreak techniques * One external partner rated it the most robust safeguards of any model tested * Mythos 5's misaligned behavior measured low, similar to Opus 4.8 There is a cost dimension too. When a request is routed to Opus 4.8, you pay Opus prices for that portion. Anthropic also requires a mandatory 30 day retention of inputs and outputs for all users, with human review capability, which it frames as a defense against novel attacks. On AWS this is enforced through a setting that must be enabled before the model can be invoked. What Mythos 5 Can Do That Fable 5 Cannot The starred rows in the benchmark table above are exactly where the two models diverge. On those, Mythos 5 answers at full capability while Fable 5's safeguards route the request to Opus 4.8, so the public model performs closer to Opus 4.8 than to the starred score. The gap is widest in the high risk domains: * Cybersecurity (ExploitBench): Mythos 5 captures 78.0% against Opus 4.8 at 40.0% and GPT-5.5 at 34.0%, nearly doubling the previous frontier * Biology hard (BioMysteryBench): Mythos 5 leads at 46.1% versus Opus 4.8 at 40.0% * Health (HealthBench Professional): Mythos 5 at 66.0% versus Opus 4.8 at 56.9% * Humanity's Last Exam with tools: 64.5% for the Mythos class versus 57.9% for Opus 4.8 A general user hitting these topics through Fable 5 is redirected to Opus 4.8, which is why public Fable 5 lands near the Opus column on the starred rows. Mythos 5, available only to vetted cyber defenders and approved biology researchers, is the version that answers in full. This is the core of Anthropic's argument for the whole release. A model this capable in cybersecurity and biology is genuinely dangerous in the wrong hands, so that capability is locked behind trusted access while Fable 5 lets everyone else use the same intelligence for everything that is not high risk. Pricing Claude Fable 5 and Claude Mythos 5 are both priced at $10 per million input tokens and $50 per million output tokens. For context, Claude Opus 4.8 costs $5 and $25, which makes Fable 5 exactly twice the price of the model just below it. Anthropic prefers to compare up rather than down. Against the earlier Claude Mythos Preview, Fable 5 costs less than half as much, so within the Mythos class the price has dropped sharply even as capability went up. Whether the premium is worth it depends on the work. For routine chat, Opus 4.8 or a cheaper tier is the obvious pick. For long autonomous jobs where a higher pass rate compounds across hundreds of steps, like the Stripe migration, the premium can pay for itself by finishing tasks a cheaper model would fail. Availability and Access Anthropic shipped Fable 5 across a wide set of platforms on launch day. * Claude API and Claude Platform, plus consumer and business subscriptions * Amazon Bedrock, live in US East (N. Virginia) and Europe (Stockholm) at launch * GitHub Copilot, generally available * Harvey, the legal AI platform There is a rollout window worth noting. Through June 22, 2026, Fable 5 is included at no extra cost on Pro, Max, Team, and seat based Enterprise plans. From June 23, access shifts to a usage credit model, so the free inclusion is effectively a two week introductory period. Claude Mythos 5 is the exception to all this breadth. It is restricted to Project Glasswing partners and a small set of approved biology researchers, and it is not part of any general subscription or public API tier. The Timing Controversy The release cannot be separated from its context. Just days before launching Fable 5, Anthropic warned that AI systems are advancing toward recursive self improvement, the point at which models can improve themselves without human intervention. The company paired that warning with a call for major AI labs to agree on a coordinated brake on frontier development. Then it shipped its most powerful model yet. The apparent contradiction was not lost on anyone covering the launch. Releasing a Mythos class model to the public, days after arguing that AI is becoming too dangerous, reads as either a reversal or a deliberate strategy. Anthropic's answer is that the safeguards are the entire point. The argument runs that a Mythos class capability is coming whether Anthropic ships it or not, so it is better for the first public model at this tier to arrive with hard guardrails, fallback routing, and mandatory monitoring than to wait for a competitor to release something equally capable with no protections. Whether that holds up is fair to debate. The safeguards are real and the red team numbers are strong, but they depend on classifiers that have to catch every dangerous request, data retention that raises privacy questions, and a fallback model users do not control. The launch is a live test of whether a model this capable can be released responsibly. Where Fable 5 Leaves Questions Open A few things are not fully nailed down yet: * Output ceiling: the 1 million token context is now corroborated by Artificial Analysis, but the rumored 128,000 token output limit is still unverified in Anthropic's materials * Vendor figures: Anthropic's benchmark numbers are its own, and the full cross model comparison across every tested category has not been published * Fallback experience: a model that silently routes 5% of sessions to an older model behaves differently near the boundary of the blocked domains, and the quality of that handoff is untested in public None of this undercuts the headline. On the benchmarks Anthropic did publish, and on Artificial Analysis's independent leaderboard, Fable 5 is the strongest model the company has released to the public. The full picture will sharpen as the model card and more independent evaluations land. Why This Release Matters Three things make this launch significant beyond the benchmark numbers. A new tier, public on day one. For two years the frontier conversation has been about incremental gains within the Opus, GPT, and Gemini flagship lines. Mythos is a deliberate step above that ladder, and Fable 5 makes it generally available rather than locking it to a research preview. A safeguard template the industry will study. Anthropic is betting that the way to ship a dangerous capability is to wrap it in classifiers, fallback routing, red teaming, and monitoring, then release the safe version broadly while the unrestricted version stays behind a trusted access wall. If it works, it becomes the playbook. If it fails, it becomes the cautionary tale. A coding gap big enough to change what agents can do. An 80.3% SWE-Bench Pro pass rate and a Stripe migration finished in a day point to a model that can carry long, complex jobs end to end rather than assisting step by step. That is the difference between a coding assistant and an autonomous engineer. For anyone who wants to see Claude together with other AI models, Fello AI puts Claude, GPT, Gemini, DeepSeek, Perplexity, and more into a single native app for Mac, iPhone, and iPad, so you can test the same task across models and pick the right one for the job without managing a stack of subscriptions.
Anthropic's Claude Fable 5 Just Launched
On June 9, 2026, Anthropic released Claude Fable 5, the first model in a brand new top tier the company calls the Mythos class. This is not another point upgrade in the Opus line. Anthropic describes Fable 5 as a model that sits above Opus in raw capability. It is state of the art on nearly every benchmark it was tested on, and it is the most powerful model the company has ever made available to the general public. It launched alongside a sibling model, Claude Mythos 5. That model is the same underlying system with the safety guardrails lifted, available only to a small set of vetted users. The new tier is the real headline. For two years Anthropic ran a three step ladder of Haiku, Sonnet, and Opus. Mythos is a fourth step above all of them, and Fable 5 is the version of that step the public can actually use. The timing is its own story. Fable 5 arrived only days after Anthropic warned that frontier AI is approaching recursive self improvement and urged the industry to agree on a coordinated brake on development. Releasing your most powerful model yet, days after that warning, struck a lot of observers as either a contradiction or a carefully chosen strategy. In this article we cover what Fable 5 is, how the Mythos class works, the full benchmark picture against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, the safeguard system, pricing, availability, and how the restricted Mythos 5 model fits in. What Claude Fable 5 Actually Is Claude Fable 5 is a frontier reasoning model and the first generally available member of Anthropic's Mythos class. Here are the basics at a glance: * API identifier: , with and on AWS * Same model as Mythos 5, separated only by a safety layer added at inference time * Built for long horizon agentic work, not quick single turn chat * State of the art vision, including reading charts and rebuilding code from screenshots * 1 million token context window, per Artificial Analysis The most important fact is that Fable 5 and Mythos 5 are the same model. The difference is not the weights or the training. It is the safety layer wrapped around Fable 5 that intercepts high risk requests. So every benchmark number Fable 5 posts is the Mythos class capability operating with a safety net underneath it. Anthropic positions the model for autonomous work that runs for hours or even days inside an agent harness. In that setting the model plans a multi step job, calls tools, reads the results, validates its own output, and corrects course without a human in the loop. The trait partners keep citing is self verification. Rakuten told Anthropic the model reflects on and validates its own work, which is what makes the autonomous operation practical rather than risky. On context size, the picture firmed up after launch. Independent benchmarking site Artificial Analysis lists a 1 million token window, matching the figure that circulated in early summaries. The companion 128,000 token maximum output number is still not confirmed in Anthropic's own materials, so treat the output ceiling as unverified for now. The Mythos Class Explained Anthropic's lineup has always been a ladder: * Haiku is the small, fast, cheap tier * Sonnet is the balanced workhorse * Opus is the frontier reasoning tier * Mythos is the new step above all three Mythos represents a capability level Anthropic had been holding back from general release, because of the risks a model that strong could pose in the wrong hands. Fable 5 is what you get when Anthropic takes a Mythos class model and adds the guardrails needed to make it safe for a broad audience. The pitch is that the safeguards are not a tax on the model. They are the thing that makes shipping it possible at all. That framing matters for the price. You are not paying twice the cost of Opus 4.8 for a slightly better Opus. You are paying for a step up to a new tier, delivered with a safety system the older tiers never needed. A Brief History of Mythos and Project Glasswing Mythos did not appear out of nowhere on launch day. The first Mythos model, Claude Mythos Preview, shipped quietly in April 2026 through a limited program named Project Glasswing. Project Glasswing was not a public product. It was a controlled access program aimed at a narrow set of users, mainly cyber defenders and critical infrastructure providers. These are organizations that could use a frontier model's offensive security knowledge to strengthen their own defenses. Anthropic used the program to learn how a Mythos class model behaves in the field before deciding whether to release it more widely. The June 9 launch is the graduation of that experiment into two official products: * Claude Mythos 5, the direct upgrade for Glasswing partners and select biology researchers, available only through trusted access * Claude Fable 5, the safeguarded version built for everyone else Anthropic has also signaled that the trusted access program for vetted cybersecurity organizations will broaden, and that enrollment for a biology research program is beginning. Benchmark Results Anthropic published a full benchmark table with the launch, putting the Mythos class against Opus 4.8, GPT-5.5, and Gemini 3.1 Pro. Launch coverage from VentureBeat and Tom's Hardware echoed the topline that it is state of the art on nearly all of them. Two things matter for reading the table. The Anthropic column combines Mythos 5 and Fable 5, showing the higher of the two, which are normally within 1 to 3 points of each other. The starred rows are the exception. On those, Fable 5's safeguards route requests to Opus 4.8, so the public Fable 5 score lands closer to Opus 4.8 while the starred figure reflects the unrestricted Mythos 5. Starred rows are where Fable 5's safeguards trigger a fallback. The number shown is Mythos 5, and public Fable 5 performs closer to Opus 4.8 on those tasks. On Terminal Bench 2.1, GPT-5.5's 83.4% is via Codex CLI and Gemini's 70.7% via Gemini CLI. The standouts are the unstarred rows, where Fable 5 and Mythos 5 are effectively the same. Fable 5 leads every rival on agentic coding, knowledge work, vision, tool use, computer use, and legal reasoning. The SWE-Bench Pro gap is the clearest signal, with Fable 5 at 80.3% sitting more than 11 points above Opus 4.8 and nearly 22 above GPT-5.5, a margin that compounds across every step of a long autonomous job. On the hardest FrontierCode Diamond split it more than doubles Opus 4.8 and roughly quintuples GPT-5.5. A few results outside the official table round out the picture: * Analytics: first model to score 90% on Hex's benchmark of complex, long running analytical tasks * Finance: highest score of any model on Hebbia's senior level finance benchmark * Memory: a persistent file based memory task improved Fable 5's performance three times more than Opus 4.8 * Spreadsheets: beats Opus 4.8 at every effort level while finishing runs 25 to 30% faster How Fable 5 Compares to Opus 4.8, GPT-5.5, and Gemini 3.1 Pro Stripping the benchmarks down to positioning, the four models line up like this. The pattern from the benchmark table is consistent. Fable 5 leads on the unsafeguarded categories, with Opus 4.8 as the closest follower. GPT-5.5 sits a step behind on coding but stays competitive on vision and spatial reasoning, and Gemini 3.1 Pro trails across the board. The tradeoff is price and the safety layer. Fable 5 costs twice what Opus 4.8 does and routes certain requests away from itself, which none of the other models here do. Independent Benchmarks from Artificial Analysis The first independent confirmation came from Artificial Analysis, which folded Fable 5 into its public leaderboard within a day of launch. Fable 5 debuted at the very top. The headline numbers: * Intelligence Index: 65, ranked number one, against a roughly 36 average for comparable models * GDPval-AA (agentic real world work): 1,932, number one, with Anthropic models taking three of the top four spots * Coding and agentic sub scores: 62 and 80.7, both at or near the top * Fallback rate: just 2% of GDPval-AA tasks deferred to Opus 4.8, matching Anthropic's under 5% claim The independent data also surfaced weaknesses the partner testimonials gloss over. The biggest one is speed: * Output speed: 60.3 tokens per second, ranked 72nd of 152 models, squarely mid pack * Time to first token: around 82 seconds, far above the peer median of under 3 seconds That latency is a direct consequence of the heavy chain of thought reasoning the model runs before answering. This is a model built for deep, long horizon work, not snappy back and forth chat. Cost is the other caveat. Artificial Analysis lists Fable 5 at the expensive end of its chart, quoting an input rate of $12.50 per million tokens against the $10 in Anthropic's own materials, with a blended rate around $8.20 once cache hits are factored in. Either way, it is among the priciest models on the board. One scoreboard is still blank. Fable 5 has not yet appeared on LMArena's Chatbot Arena leaderboard, which ranks models by head to head human preference votes. That is expected for a model this new, since Arena needs a large volume of comparisons before assigning a reliable rating. A human preference ranking will be one of the more interesting data points to watch in the coming weeks. Real World Performance from Launch Partners Benchmarks are one signal. The launch partner reports point in the same direction. Stripe produced the most cited result. Anthropic says Fable 5 compressed months of engineering into days. The specific example is a migration across a 50 million line Ruby codebase that the model completed in a single day, work that would otherwise have taken a full engineering team over two months. Rakuten emphasized reliability over raw speed. The company reported that Fable 5 reflects on and validates its own work, letting it run autonomous operations where the model is trusted to check itself rather than handing every step back to a human. Hex contributed the 90% analytics milestone, the first time any model cleared that bar on its suite of complex tasks. AWS framed the model as purpose built for long running, asynchronous execution, the kind of job that can run for days inside a harness before producing a result. The Safeguard System The safeguards are what make Fable 5 a public product, so they deserve a close look. Fable 5 ships with classifiers covering four high risk domains. When a request trips one, the model blocks its own response and falls back to Claude Opus 4.8, the older and more conservative model, to handle the query safely. The domains are: * Cybersecurity, blocking exploitation and offensive cyber tasks * Biology and chemistry, blocking risky dual use research * Distillation, blocking attempts to extract or copy the model's capabilities The key number is how often this happens. Anthropic says more than 95% of Fable 5 sessions involve no fallback at all. Only around 5% hit a classifier and defer to Opus 4.8, so for the overwhelming majority of normal use you are talking to the full Mythos class model. The red team results are the other half of the safety case: * No universal jailbreaks across more than 1,000 hours of external bug bounty testing * Zero harmful single turn completions on cyberattack planning across 30 jailbreak techniques * One external partner rated it the most robust safeguards of any model tested * Mythos 5's misaligned behavior measured low, similar to Opus 4.8 There is a cost dimension too. When a request is routed to Opus 4.8, you pay Opus prices for that portion. Anthropic also requires a mandatory 30 day retention of inputs and outputs for all users, with human review capability, which it frames as a defense against novel attacks. On AWS this is enforced through a setting that must be enabled before the model can be invoked. What Mythos 5 Can Do That Fable 5 Cannot The starred rows in the benchmark table above are exactly where the two models diverge. On those, Mythos 5 answers at full capability while Fable 5's safeguards route the request to Opus 4.8, so the public model performs closer to Opus 4.8 than to the starred score. The gap is widest in the high risk domains: * Cybersecurity (ExploitBench): Mythos 5 captures 78.0% against Opus 4.8 at 40.0% and GPT-5.5 at 34.0%, nearly doubling the previous frontier * Biology hard (BioMysteryBench): Mythos 5 leads at 46.1% versus Opus 4.8 at 40.0% * Health (HealthBench Professional): Mythos 5 at 66.0% versus Opus 4.8 at 56.9% * Humanity's Last Exam with tools: 64.5% for the Mythos class versus 57.9% for Opus 4.8 A general user hitting these topics through Fable 5 is redirected to Opus 4.8, which is why public Fable 5 lands near the Opus column on the starred rows. Mythos 5, available only to vetted cyber defenders and approved biology researchers, is the version that answers in full. This is the core of Anthropic's argument for the whole release. A model this capable in cybersecurity and biology is genuinely dangerous in the wrong hands, so that capability is locked behind trusted access while Fable 5 lets everyone else use the same intelligence for everything that is not high risk. Pricing Claude Fable 5 and Claude Mythos 5 are both priced at $10 per million input tokens and $50 per million output tokens. For context, Claude Opus 4.8 costs $5 and $25, which makes Fable 5 exactly twice the price of the model just below it. Anthropic prefers to compare up rather than down. Against the earlier Claude Mythos Preview, Fable 5 costs less than half as much, so within the Mythos class the price has dropped sharply even as capability went up. Whether the premium is worth it depends on the work. For routine chat, Opus 4.8 or a cheaper tier is the obvious pick. For long autonomous jobs where a higher pass rate compounds across hundreds of steps, like the Stripe migration, the premium can pay for itself by finishing tasks a cheaper model would fail. Availability and Access Anthropic shipped Fable 5 across a wide set of platforms on launch day. * Claude API and Claude Platform, plus consumer and business subscriptions * Amazon Bedrock, live in US East (N. Virginia) and Europe (Stockholm) at launch * GitHub Copilot, generally available * Harvey, the legal AI platform There is a rollout window worth noting. Through June 22, 2026, Fable 5 is included at no extra cost on Pro, Max, Team, and seat based Enterprise plans. From June 23, access shifts to a usage credit model, so the free inclusion is effectively a two week introductory period. Claude Mythos 5 is the exception to all this breadth. It is restricted to Project Glasswing partners and a small set of approved biology researchers, and it is not part of any general subscription or public API tier. The Timing Controversy The release cannot be separated from its context. Just days before launching Fable 5, Anthropic warned that AI systems are advancing toward recursive self improvement, the point at which models can improve themselves without human intervention. The company paired that warning with a call for major AI labs to agree on a coordinated brake on frontier development. Then it shipped its most powerful model yet. The apparent contradiction was not lost on anyone covering the launch. Releasing a Mythos class model to the public, days after arguing that AI is becoming too dangerous, reads as either a reversal or a deliberate strategy. Anthropic's answer is that the safeguards are the entire point. The argument runs that a Mythos class capability is coming whether Anthropic ships it or not, so it is better for the first public model at this tier to arrive with hard guardrails, fallback routing, and mandatory monitoring than to wait for a competitor to release something equally capable with no protections. Whether that holds up is fair to debate. The safeguards are real and the red team numbers are strong, but they depend on classifiers that have to catch every dangerous request, data retention that raises privacy questions, and a fallback model users do not control. The launch is a live test of whether a model this capable can be released responsibly. Where Fable 5 Leaves Questions Open A few things are not fully nailed down yet: * Output ceiling: the 1 million token context is now corroborated by Artificial Analysis, but the rumored 128,000 token output limit is still unverified in Anthropic's materials * Vendor figures: Anthropic's benchmark numbers are its own, and the full cross model comparison across every tested category has not been published * Fallback experience: a model that silently routes 5% of sessions to an older model behaves differently near the boundary of the blocked domains, and the quality of that handoff is untested in public None of this undercuts the headline. On the benchmarks Anthropic did publish, and on Artificial Analysis's independent leaderboard, Fable 5 is the strongest model the company has released to the public. The full picture will sharpen as the model card and more independent evaluations land. Why This Release Matters Three things make this launch significant beyond the benchmark numbers. A new tier, public on day one. For two years the frontier conversation has been about incremental gains within the Opus, GPT, and Gemini flagship lines. Mythos is a deliberate step above that ladder, and Fable 5 makes it generally available rather than locking it to a research preview. A safeguard template the industry will study. Anthropic is betting that the way to ship a dangerous capability is to wrap it in classifiers, fallback routing, red teaming, and monitoring, then release the safe version broadly while the unrestricted version stays behind a trusted access wall. If it works, it becomes the playbook. If it fails, it becomes the cautionary tale. A coding gap big enough to change what agents can do. An 80.3% SWE-Bench Pro pass rate and a Stripe migration finished in a day point to a model that can carry long, complex jobs end to end rather than assisting step by step. That is the difference between a coding assistant and an autonomous engineer. For anyone who wants to see Claude together with other AI models, Fello AI puts Claude, GPT, Gemini, DeepSeek, Perplexity, and more into a single native app for Mac, iPhone, and iPad, so you can test the same task across models and pick the right one for the job without managing a stack of subscriptions.
Perplexity plans 2028 IPO regardless of Anthropic or OpenAI listings, CNBC reports
June 8 (Reuters) - AI firm Perplexity is planning to go public in 2028 regardless of how the market receives the listings of Anthropic and OpenAI, CNBC reported on Monday, citing an interview with CEO Aravind Srinivas. o "Agnostic of these two companies, we were planning for something in 2028, so that still remains the case," Srinivas told CNBC in an interview. o OpenAI confidentially filed for a U.S. IPO earlier on Monday, following Anthropic's filing last week. Elon Musk's SpaceX is also preparing to go public on Friday. o "I certainly think there will be ripple effects if they don't go well, like there is no sugar coating on that. The SpaceX IPO this week will definitely be a leading indicator of how Anthropic or OpenAI will go out," Srinivas told CNBC. o "I think it's important for the AI industry that these IPOs go well, and I actually think they will go well, because they're doing well," Srinivas added. o In 2025, addressing speculation about Perplexity's finances, Srinivas said the company was not running out of money and had no plans to go public before 2028. (Reporting by Natalia Bueno Rebolledo in Mexico City; Editing by Rashmi Aich)
Perplexity plans 2028 IPO regardless of Anthropic or OpenAI listings, CNBC reports
June 8 (Reuters) - AI firm Perplexity is planning to go public in 2028 regardless of how the market receives the listings of Anthropic and OpenAI, CNBC reported on Monday, citing an interview with CEO Aravind Srinivas. * "Agnostic of these two companies, we were planning for something in 2028, so that still remains the case," Srinivas told CNBC in an interview. * OpenAI confidentially filed for a U.S. IPO earlier on Monday, following Anthropic's filing last week. Elon Musk's SpaceX is also preparing to go public on Friday. * "I certainly think there will be ripple effects if they don't go well, like there is no sugar coating on that. The SpaceX IPO this week will definitely be a leading indicator of how Anthropic or OpenAI will go out," Srinivas told CNBC. * "I think it's important for the AI industry that these IPOs go well, and I actually think they will go well, because they're doing well," Srinivas added. * In 2025, addressing speculation about Perplexity's finances, Srinivas said the company was not running out of money and had no plans to go public before 2028. (Reporting by Natalia Bueno Rebolledo in Mexico City; Editing by Rashmi Aich)
Perplexity planning IPO in 2028 regardless of what happens to Anthropic or OpenAI, CEO tells CNBC By Reuters
Risk Disclosure: Trading in financial instruments and/or cryptocurrencies involves high risks including the risk of losing some, or all, of your investment amount, and may not be suitable for all investors. Prices of cryptocurrencies are extremely volatile and may be affected by external factors such as financial, regulatory or political events. Trading on margin increases the financial risks. Before deciding to trade in financial instrument or cryptocurrencies you should be fully informed of the risks and costs associated with trading the financial markets, carefully consider your investment objectives, level of experience, and risk appetite, and seek professional advice where needed. Fusion Media would like to remind you that the data contained in this website is not necessarily real-time nor accurate. The data and prices on the website are not necessarily provided by any market or exchange, but may be provided by market makers, and so prices may not be accurate and may differ from the actual price at any given market, meaning prices are indicative and not appropriate for trading purposes. Fusion Media and any provider of the data contained in this website will not accept liability for any loss or damage as a result of your trading, or your reliance on the information contained within this website. It is prohibited to use, store, reproduce, display, modify, transmit or distribute the data contained in this website without the explicit prior written permission of Fusion Media and/or the data provider. All intellectual property rights are reserved by the providers and/or the exchange providing the data contained in this website. Fusion Media may be compensated by the advertisers that appear on the website, based on your interaction with the advertisements or advertisers.
Perplexity planning IPO in 2028 regardless of what happens to Anthropic or OpenAI, CEO tells CNBC
June 8 (Reuters) - AI firm Perplexity is planning to go public in 2028 regardless of how the market receives the listings of Anthropic and OpenAI, CNBC reported on Monday, citing an interview with CEO Aravind Srinivas. "Agnostic of these two companies, we were planning for something in 2028, so that still remains the case," Srinivas told CNBC in an interview. (Reporting by Natalia Bueno Rebolledo in Mexico City; Editing by Rashmi Aich)
ChatGPT vs Gemini vs Claude vs Perplexity: The 2026 AI Engine Landscape
Updated June 2026. Originally published 2023 when Bard launched, refreshed for the AI Communications era -- including Bard's rebrand to Gemini, the rise of Claude and Perplexity, and the structural shift to AI Overviews. Gemini (formerly Bard) vs ChatGPT vs the Modern AI Engine Landscape When this piece was originally published in May 2023, the AI chatbot landscape was simple: ChatGPT had taken the internet by storm, and Google had just launched Bard as its first answer. Three years later the landscape looks completely different. Google Bard has been rebranded to Gemini and integrated across Google's product suite. Anthropic's Claude has emerged as one of the most-used AI engines among professionals and developers. Perplexity has become the dominant answer-engine for citation-driven research. ChatGPT itself has gone through multiple GPT model generations. And the most important structural change -- Google AI Overviews now mediates a substantial share of all U.S. search queries with AI-generated answers above the traditional ten blue links. This piece compares the modern AI engine landscape and what it means for marketers, communicators, and brands operating in the answer-engine era. The Five Major AI Engines in 2026 ChatGPT (OpenAI) Launched November 2022, ChatGPT remains the largest single AI chatbot by consumer adoption. The product has gone through GPT-3.5, GPT-4, GPT-4o, and subsequent model generations. ChatGPT now offers web browsing, file uploads, image generation, voice mode, and integration with custom tools. The user base spans casual consumers, students, professionals, and developers. ChatGPT is generally regarded as strong on writing, summarization, brainstorming, and conversational tasks. Gemini (Google, formerly Bard) Google rebranded Bard to Gemini in February 2024 and has continued to expand the product across Google's surfaces -- the standalone Gemini app, the AI Overviews appearing in Google Search results, the AI-powered features in Gmail, Docs, Sheets, and the broader Workspace suite, and the Android assistant integration. Gemini's strengths reflect Google's data and infrastructure advantages: real-time web access, multimodal capability, integration with the broader Google ecosystem, and the unique distribution that places AI answers in front of billions of daily searchers. Claude (Anthropic) Anthropic's Claude has emerged as one of the most-used AI engines among professionals, developers, and enterprise users. Claude is generally regarded as strong on long-form writing, complex reasoning, coding, and tasks requiring careful instruction-following. The user base skews professional and technical relative to ChatGPT's broader consumer reach. Claude is available through the standalone Claude app, the API, and integration with platforms like Slack and increasingly the broader enterprise software ecosystem. Perplexity Perplexity launched as an "answer engine" -- an AI-powered search experience that returns synthesized answers with explicit citations to the sources. The product has become the dominant tool for citation-driven research, professional information work, and any use case where users need to verify the source of an AI answer. Perplexity's growth has been particularly strong among researchers, journalists, financial analysts, and other professionals whose work depends on traceable sourcing. Google AI Overviews Not a chatbot but the most consequential AI engine in the search market. AI Overviews are the Google-generated AI summaries that now appear above traditional search results on a substantial share of U.S. queries. For brands, AI Overviews are the single most important AI surface to optimize for, because they intercept buyer research at the moment of intent. The brand mentioned inside the AI Overview wins the consideration. The brand omitted is invisible. The Original Comparison -- What Changed Since 2023 When this piece was first written, the comparison was simpler. ChatGPT was trained on data up to 2021 and could not browse the web. Bard used Google's real-time web access. ChatGPT used GPT-3.5 (with GPT-4 in the paid tier). Bard used PaLM 2. Both were "in the initial phases of learning." The structural changes since: * Real-time web access is now standard across all major AI engines, not a Bard differentiator. * Bard became Gemini and the underlying model migrated from PaLM 2 through the Gemini model family. * ChatGPT's training data window moved forward through multiple model generations. * The "summarizing and writing vs. comprehensive answers" distinction has largely collapsed -- all major AI engines now do both well. * The arrival of Claude and Perplexity changed the landscape from a two-horse race into a multi-engine ecosystem where buyers and professionals use different engines for different tasks. * The arrival of AI Overviews made the comparison less about chatbot UX and more about which AI surfaces mediate buyer research. What This Means for Brands and Communicators Three implications: 1. Citation Share across multiple engines is the new metric. Brands need to measure their appearance inside ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews -- not just one engine. The brand that appears inside three of five engines is differently positioned than the brand that appears inside only one. 2. Generative Engine Optimization (GEO) is the new discipline. The optimization work that made content findable in Google has evolved. GEO is the practice of structuring content, building editorial authority, and creating the citation patterns that AI engines reward. 3. AI Communications is the new function. Public relations, digital marketing, GEO, and AI-visibility research now combine into a single discipline. The firms and in-house teams that operate across this stack are pulling ahead of teams still organized around the pre-AI playbook. Other AI Engines and Tools Worth Knowing Beyond the five major engines above, the AI tool landscape includes specialist products (Jasper for marketing content, Cursor and GitHub Copilot for code, Midjourney and DALL-E for image generation), open-source models (Meta's Llama family, Mistral, others), enterprise-focused offerings, and the Chinese AI ecosystem (Baidu's Ernie, ByteDance's Doubao, DeepSeek). For Western brands building AI Communications strategies, the five major engines covered above are the priority. The specialist and open-source tooling supports the workflow but does not currently mediate buyer research at scale. Frequently Asked Questions Is Google Bard the same as Gemini? Yes. Google rebranded Bard to Gemini in February 2024 and consolidated its consumer and enterprise AI offerings under the Gemini name. The underlying model and the product surface have continued to evolve. Which AI engine is best for which task? ChatGPT remains the leading consumer AI for general use. Claude is preferred by many professionals for long-form writing and complex reasoning. Perplexity is the dominant citation-driven research tool. Gemini wins on Google ecosystem integration and AI Overviews distribution. The right answer depends on the task -- and most professionals now use multiple engines for different jobs. What is the most important AI engine for brand visibility? Google AI Overviews, because of its reach inside Google Search where most consumer and B2B research still begins. ChatGPT, Claude, Perplexity, and Gemini are also important -- measurement and optimization should cover all five for any brand serious about AI Communications. How should marketers prepare for the AI Communications era? Build Citation Share measurement across the five major engines, develop Generative Engine Optimization (GEO) capability, accumulate editorial authority through original reporting and named expert voices, and reorganize the marketing and communications function around AI Communications as an integrated discipline rather than treating AI as a content-generation tool inside the existing function.
CNN Sues Perplexity For Alleged Copyright Infringement
While other MediaPost newsletters and articles remain free to all ... our new Research Intelligencer service is reserved for paid subscribers ... Subscribe today to gain access to every Research Intelligencer article we publish as well as the exclusive daily newsletter, full access to The MediaPost Cases, first-look research and daily insights from Joe Mandese, Editor in Chief.
CNN sues Perplexity AI over alleged scraping of 17,000 news stories
CNN's lawsuit against Perplexity, OpenAI's EU compliance framework, and DOJ's intervention in Colorado's AI Act all landed on same day On a single day, Thursday, a renowned media outlet, CNN, filed a copyright lawsuit against Perplexity AI; OpenAI published a formal internal governance framework aligned to EU and California law; and the US Department of Justice filed its first-ever federal challenge to a state AI statute in Colorado. When all three cases were put together, they represented something the industry wanted to avoid. It was fragmentation of artificial intelligence regulations in courts, businesses, and governments, with no one solution gaining any traction. Why are publishers suing Perplexity? According to CNN's filing of a 54-page case against Perplexity, a new AI firm, the plaintiff claims that the company scraped and distributed over 17,000 pieces of their articles, photographs, and videos to use this information as input into AI-based answers that, according to CNN, are competing with their news content. CNN alleges Perplexity falsely implied a content relationship by advertising CNN premium access to subscribers of its Comet Plus tier, despite no licensing agreement existing between the two companies. Negotiations between the media organisation and Perplexity had been made prior to the lawsuit being filed. Perplexity Chief Communications Officer Jesse Dwyer gave the following response from the company, according to which 'You can't copyright facts.' On the other hand, CNN's claim is based on the infringement of copyrights on works containing protected expression, such as articles, photos, and videos. Other news agencies such as Time, Gannett, Le Monde, and Der Spiegel chose to license Perplexity instead of suing. However, as time passes by, the price for legally licensed copyrighted training data grows with each new legal case. As reported in relation to Bartz v. Anthropic, a class action suit that saw the plaintiffs claim that their copyrighted books were used to train AI models without their consent, a settlement worth billions is awaiting the court's approval in mid-May 2026 in the Northern District of California.
How Perplexity's 'Personal Computer' Marketing is Turning Apple's Mac Mini Into the Hottest Hardware in AI
The scramble to secure computing power for artificial intelligence is no longer confined to billion-dollar data centers and advanced Nvidia chips. Increasingly, it is spilling into the consumer hardware market, where an unlikely device has emerged as a favorite among AI developers, power users and technology enthusiasts: Apple's Mac Mini. That trend is now being amplified by Perplexity, which has been sending Mac Minis to select users as part of an effort to showcase its new AI agent platform, Personal Computer. A handful of technology-focused creators began posting online in recent weeks that they had received Mac Minis from Perplexity. The company later confirmed it had distributed a small number of devices to people interested in exploring the full capabilities of Personal Computer, its newest push beyond AI search and into autonomous digital assistants. The campaign may appear at first glance to be a straightforward influencer-marketing exercise. In reality, it highlights a much larger shift underway in the AI industry: the growing importance of personal computing infrastructure as AI agents become capable of performing increasingly complex tasks on behalf of users. Perplexity's Personal Computer, which began rolling out in April, expands on the company's browser-based AI agent technology by allowing AI to operate across local files, native applications, and web services. Unlike traditional chatbots that simply answer questions, the system is designed to interact directly with a user's digital environment. Perplexity has described the Mac Mini as "one of the best ways to experience Personal Computer." "On a mini, Personal Computer stays available 24/7 for work that needs a persistent machine or secure local access to your files and native apps," the company wrote in an April blog post. That statement offers a glimpse into where AI development is headed. For years, most advanced AI services have relied almost entirely on cloud computing. Users submit requests to remote servers, which process information and return answers. But as AI agents evolve from conversational tools into software capable of carrying out multi-step actions, there is a growing demand for systems that remain continuously available, retain access to local files, and operate with lower latency. The Mac Mini has emerged as an attractive platform for that transition. Its appeal stems from a combination of factors: powerful Apple silicon processors, relatively low energy consumption, quiet operation, and the ability to remain online continuously. Those characteristics make it particularly suitable for running AI agents that need persistent access to applications, documents, and workflows. Perplexity Chief Communications Officer Jesse Dwyer underscored that use case, telling Business Insider that he uses his Mac Mini constantly and accesses it remotely through other Apple devices regardless of location. The enthusiasm around the machine is not limited to Perplexity. Across the AI ecosystem, developers and hobbyists have increasingly embraced the Mac Mini as a dedicated AI workstation capable of running agent-based systems, coding assistants, and local AI applications. What was once considered Apple's most overlooked desktop product is increasingly being viewed as an affordable gateway into AI-powered productivity. The surge in interest has become significant enough to affect supply. During a March earnings call, Tim Cook highlighted strong demand for the product. Availability has tightened in recent weeks as more consumers, developers, and AI enthusiasts seek out the device. Apple's base-model Mac Mini has become increasingly difficult to find, leaving many buyers with higher-priced configurations. The phenomenon illustrates a broader shift in how AI is being commercialized. Much of Wall Street's attention remains focused on companies building large language models, AI chips, and cloud infrastructure. Yet a parallel market is emerging around the hardware required to run AI agents in everyday environments. As those systems become more autonomous, users may want dedicated machines that function as personal AI hubs. That could create new opportunities not only for Apple but also for software companies seeking to build ecosystems around AI-native computing. For Perplexity, this strategy also means expanding beyond its roots as an AI-powered search engine into a broader platform designed to compete for users' daily workflows. By promoting Personal Computer through influential technology creators, the company is attempting to position itself at the center of the emerging agent economy. The move comes as rivals race to build similar capabilities. Companies such as OpenAI, Anthropic, and Google are all developing increasingly sophisticated AI agents capable of carrying out tasks rather than merely generating responses.
Perplexity AI Says 'You Can't Copyright Facts' in Defense Against CNN Copyright Suit
More than 100 copyright lawsuits have been filed against AI companies as of early 2026. The television network CNN is taking aim at the artificial intelligence search engine Perplexity in a lawsuit over copyright infringement. As reported by the network's Brian Stetler, the suit, filed Thursday in a New York District Court, accuses the AI company of copying and distributing CNN's content, including over 17,000 of CNN's stories, videos, images and other published works. Though this is CNN's first legal case against an AI company, the network joins other publishers who have sued the San Francisco-based startup, including the New York Times and News Corp. According to the suit, CNN attempted to strike a licensing deal with Perplexity, but those talks didn't result in an agreement. CNN previously made a content licensing deal with Meta last year, where the tech giant compensates the media company for using its reporting and content to respond to queries on Meta AI. AI products regularly scrape news publications and websites to answer user questions with real-time data, accelerating the collapse in traffic and revenue to original sources. In response to the lawsuit, Jesse Dwyer, Perplexity's chief communications officer, told Stetler and other media outlets in a statement: "You can't copyright facts." The US government's Copyright Office states: "Copyright does not protect facts, ideas, systems, or methods of operation, although it may protect the way these things are expressed." CNN said in its own statement that a company valued at tens of billions of dollars shouldn't "steal from entities that create the original content Perplexity exploits" and that "commercial operators can and must pay to make use of it." A Perplexity representative didn't immediately respond to a request for comment. AI copyright suits Perplexity is one of several companies, including OpenAI and Anthropic, that have been battling news publishers and media giants over copyright claims. (Disclosure: Ziff Davis, CNET's parent company, in 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.) More than 100 such lawsuits have been filed. But different conclusions have been reached as to whether training AI models on copyrighted data counts as fair use, said Michael Goodyear, an associate professor at New York Law School. Considerations include how the training occurs, what AI outputs contain and whether there's any competitive harm to copyright holders. "No appellate courts have yet weighed in on the viability of these copyright infringement claims against AI companies," Goodyear said. In the CNN case, he said that Perplexity is correct that facts aren't protected by copyright, but the way CNN presents facts could be. "Even short news articles would typically qualify for copyright protection under the low bar of required originality," Goodyear said. "The question becomes whether the thousands of cases of infringement CNN describes are copying whole paragraphs verbatim, or whether they are paraphrasing or merely copying unprotectable facts." AI licensing deals As plunging website traffic has drained billions in publisher revenue and triggered widespread media layoffs, AI firms are aggravating the crisis. According to a new report from the think tank Open Markets Institute, over the past six months, the rate of AI crawlers bypassing paywalls and blocks has nearly quadrupled, spiking from 3.3% to 12.9%. That's partly why a number of publishers signed AI content licensing deals with tech companies to monetize content used to train AI systems. One way out for Perplexity may be to renegotiate a licensing deal with CNN. Even if Perplexity has valid legal arguments, a licensing agreement could shift from unauthorized scraping toward a formalized content partnership. However, the Open Markets Institute report says that when it comes to AI content licensing, news and content creators are trapped in a double bind. The same tech giants whose AI tools are starving websites of human traffic are now the ones gatekeeping the licensing deals meant to replace that lost ad revenue.
CNN Slaps Perplexity AI With New Copyright Lawsuit | Law.com
Brooklyn Federal Judge Rejects Hobbs Act Robbery as Predicate for 'Career Criminal' Enhanced Sentence U.S. District Court Judge Eric Vitaliano of the Eastern District of New York ruled that Hobbs Act robbery does not qualify as a "violent felony" under the Armed Career Criminal Act, finding the offense too broad under the categorical approach because it can be committed against property. The U.S. Court of Appeals for the Second Circuit has not yet weighed in on this issue.
I paid for Perplexity, Claude, and NotebookLM, and only one was worth keeping
There's a point in AI adoption where the tools stopped being optional. Think about it, how many tools have you tried out at this point or even become reliant on, spanning across domains like productivity, creativity, coding, learning, and so on - a big chunk of this stuff happens in some kind of chat window now. This shift toward AI isn't really news anymore, but with so many new companies popping up all wanting you to try their product, the bill is. And it's not like AI subscriptions are $5 throwaways - some are $20 a pop, if not more, and they all sit alongside whatever else you've already got renewing in the background. Every single AI tool is competing for the same slot in your monthly budget. So the key is to pick only what you actually need, and stick to the free tiers on the rest. At least, that's how I've been approaching it. And that's why I gave three different categories a fair shot: a search engine, source-grounded research, and a general chatbot. I subscribed to Claude, Perplexity, and NotebookLM (Google AI), and here's how that panned out for me... Want to stay in the loop with the latest in AI? The XDA AI Insider newsletter drops weekly with deep dives, tool recommendations, and hands-on coverage you won't find anywhere else on the site. Subscribe by modifying your newsletter preferences! I use ChatGPT, Claude, Perplexity, and Gemini daily -- here's the only one worth paying for One stands above the rest. Posts 16 By Mahnoor Faisal The case for Perplexity Pro My other search engines were already giving me what I needed for free First up is Perplexity. The Pro tier is $20 a month, and it's the obvious upgrade pitch for anyone who already uses the free version a lot. You get unlimited Pro Searches instead of being capped daily, access to bigger models like GPT-5, Claude Sonnet 4.6, and Gemini 3 Pro instead of being routed through whatever default it gives you, way more file uploads inside Spaces (50 per Space instead of 5), and Deep Research without the monthly limit you hit pretty quickly on the free plan. There's also image generation, some API credits, and access to a few premium data sources like Statista that would normally cost money to read. So on paper, I was getting plenty. And I do genuinely like Perplexity; the citation system is one of the better things going for it, and Deep Research is great when you actually need a structured multi-source breakdown of something instead of just an answer. Spaces also has potential as a sort of project hub for research. But after the month was over, I didn't feel like I was actually getting anything that I couldn't already get by combining my existing search engines and chatbots. Both Google and Brave have some AI features baked in already, their AI summaries and dedicated chatbots have gotten really good over the past year or so. So the $20 didn't earn its place for me, however, if your stack is research-heavy, I would actually recommend Perplexity Pro. Perplexity See at Perplexity Expand Collapse Claude Pro's turn You don't need to be a coder to get your money's worth Pro is also $20 a month, and you get a lot for it - the obvious thing being five times the usage of the free tier, but the bigger draw is access to Opus 4.7 (Anthropic's most capable model) and Sonnet 4.6, both with a 200k context window that you can push up to 1 million tokens with Extra Usage enabled. You also get unlimited Projects instead of being capped at 5, Artifacts with way more headroom, the inline interactive visuals, Claude Design, and Cowork. Claude Code is also included but it's not something I need for my workflow as of yet. The thing about Opus 4.7 is that it actually pushes back, which is a weirdly rare thing in chatbots right now. Most of them are just trying to agree with you faster. Opus will tell me when an approach isn't going to work, and that makes it a much better thinking partner for the design work and research I do. Projects is still the feature I lean on the hardest - my design briefs, novel drafts, and research stacks all live in their own contained spaces, which means no re-explaining the context every single time. Claude Design has also been a nice addition for quick visual mockups and honestly just playing around with UI concepts. Plus, I've got Cowork set up to handle a couple of folders in my Obsidian vault and screenshots directory on a schedule. Usage caps still hit on the heaviest design days, but most of the time the sessions stay smooth. So $20 for Opus, Artifacts, Projects, inline visuals, Claude Design, and Cowork - all wrapped in an interface that's mostly pleasant to use - is the easiest subscription justification in my stack. It's truly an all-in-one tool, it spans across so many domains, and it's why I've actually had the subscription for three months now. I cancelled my ChatGPT, Perplexity, and Gemini subscriptions for Claude -- and I should have sooner Wish I did this sooner. Posts 50 By Mahnoor Faisal Claude OS Windows, macOS Individual pricing Free plan available; $17/month Pro plan Group pricing $100/month per person for the Max plan See at Claude Expand Collapse And then there's NotebookLM The cheapest option, with unexpected benefits NotebookLM is the one I'm newest to in subscription form. You can't actually buy NotebookLM Plus on its own - it's bundled into Google AI Plus, which is $8 a month and is the cheapest of the three. The plan bumps up NotebookLM's limits across the board: 200 notebooks instead of 100, 100 sources per notebook instead of 50, 200 daily chat queries instead of 50, more Audio and Video Overviews per day, and 3 Deep Research reports daily instead of 10 a month. Basically every limit roughly doubles. I haven't been on it for a full month yet, so the verdict is still a bit early. But the source bump alone has been the most useful thing for me - my notebooks already hit the free tier ceiling pretty quickly, so being able to load more in without juggling them has been worth it on its own. What I didn't consider was the Gemini side of things: Google AI Plus also gives you access to Gemini 3 Pro in the Gemini app with a bigger context window and higher daily prompt limits, so my Gemini chats feel noticeably more capable and run longer. If Claude wasn't a thing, Google AI is probably the subscription I would have swung toward because Gemini is a close second for me. Who knows, I might even renew it next month. NotebookLM See at NotebookLM Expand Collapse It honestly just depends on what you actually need There's no universally right answer here. If your work is research-heavy, I'd probably go with Perplexity. NotebookLM is a steal if you're already deep into the Google ecosystem and use NotebookLM regularly. Claude is the one that does the most across the widest range of work. For me, Claude is the one I'm keeping without a doubt because of its models, the design workflows, and projects. I will say, Gemini is actually better with live web data than Claude, so I'm actually tempted to keep the Google AI sub and see where it goes, but the free tier has been serving me well too. At the end of the day, it entirely depends on your workflow and preferences. I just wouldn't recommend stacking three or more of them just because the hype says you should. Figure out what kind of AI work you actually do most, and pay for that one, if the free tier has too many barriers.
