Developers logged into Anthropic's site this week and saw it: Claude Code, the prized coding tool, vanished from the Pro subscription list. No email. No heads-up. Just an "X" where a check mark once sat on the pricing page.

Spotlight first hit when AI skeptic Ed Zitron shared screenshots. By Tuesday, the phrase "includes Claude Code" was gone. The Register confirmed the shift, noting inconsistencies -- some pages still listed Pro access, while updated docs pinned it to the pricier Max plan.

Claude.ai itself, when prodded, claimed Pro users still get it. But reality bit harder. Developers vented on Reddit and X. One post: "Anthropic removed Claude Code from Pro plan with zero announcement." (@hrshkshri). Another: "Feels like they're getting desperate... Why is Anthropic the only company going this hard to milk their users?" (@nateless).

Backlash built fast. And Anthropic blinked.

Head of growth Amol Avasare jumped on X. "For clarity, we're running a small test on ~2 percent of new prosumer signups," he posted. "Existing Pro and Max subscribers aren't affected." (link). He promised: existing users would hear directly, not via "a screenshot on X or Reddit."

Anthropic's capacity crunch forces the hand.

Here's the rub. Subscriptions like Pro -- at $20 a month -- charge far below the token costs they guzzle. Sometimes by tenfold, per industry watchers. Claude's exploded in use since bundling Code into Max post-Opus 4. Add Cowork. Long-running agents. Engagement per user? Skyrocketing.

Plans weren't built for this. Avasare explained: "When we launched Max a year ago, it didn't include Claude Code, Cowork didn't exist, and agents that run for hours weren't a thing. Max was designed for heavy chat usage, that's it." Now? They're testing tweaks -- weekly caps, peak limits -- to match reality.

Last month brought off-peak nudges, utility-style. Capacity woes plague peers too. GitHub grounded Copilot accounts (The Register). Google hit similar walls.

But why test publicly? Without fanfare? A spokesperson dodged that one. Sites stayed messy -- the Claude Code page still touted Pro access. CLI output? Still said "Claude Pro." Confusion reigned.

Developers smell a push to Max -- $100 to $200 tiers. Or cheaper rivals: China's Minimax, Qwen, Kimi, GLM. "Many are already eyeing Kimi for better value in coding," noted one X user (@MikelEcheve). Risky. Trust erodes fast in AI.

Enterprise eyes watch closer. They demand stability. Sudden shifts? Uncertainty. Anthropic's pattern stings more amid prior moves. Early April, they axed Pro/Max for third-party agents like OpenClaw -- straining compute, per head of Claude Code Boris Cherny (VentureBeat, April 3). "Subscriptions weren't built for the usage patterns of these third-party tools," he wrote.

That cutoff hit April 4, 12 p.m. PT. Less than 24 hours' notice. OpenCode users got 403 errors. Loopholes closed -- Pro keys were subsidized gold for heavy API-like burns.

Devs grumbled then too. Hacker News threads lit up: "It was a loophole. They closed it." But others fumed over lost tooling freedom.

History repeats. January blocks on third-party harnesses spoofing Claude Code (VentureBeat). Revocations to rivals like xAI, OpenAI. Leaks? 8,000 GitHub repos yanked in April cleanup gone awry (TechCrunch).

So what's next? Avasare hints at options. Feedback flows. But silence on rollout irks. Pro users hold -- for now. New signups? Guinea pigs in a 2% test.

One dev quipped: "Time to try codex... Or coding manually." (@glennotiende). Others cancel. Churn brews.

Anthropic bets on Claude's pull -- top coding agent, per fans. But nickel-and-diming? In a field of alternatives? Bold. Or brittle.

Watch the pricing page. Changes stick there first. And developers talk. Loudly.

Anthropic's Stealth Test Yanks Claude Code from Pro Plans, Sparking Developer Fury