Anthropic launches Claude Opus 4.7 with improved reasoning, coding and vision capabilities

One of the key upgrades in Opus 4.7 is its ability to handle extended, multi-step workflows with greater consistency.

Anthropic has launched Claude Opus 4.7, a new version of its large language model designed to bridge the gap between its production-ready systems and the more restricted "Mythos Preview" model.

The company said Opus 4.7 is not as broadly capable as Mythos but incorporates several improvements aimed at real-world deployment, particularly in areas such as long-running task execution, structured reasoning and instruction adherence.

One of the key upgrades in Opus 4.7 is its ability to handle extended, multi-step workflows with greater consistency. The model is designed to support long-running coding and agent-based tasks, maintaining context over longer sessions and reducing the need for repeated user intervention.

Also read: Google's 99% ad safety claim leaves advertisers questioning the cost of the remaining 1%

It also introduces improved reasoning capabilities, including the ability to break down complex problems into sequential steps and verify its own outputs. This self-checking mechanism is aimed at reducing errors during extended tasks and improving reliability.

In software engineering use cases, Opus 4.7 demonstrates stronger performance in handling complex coding problems compared to its predecessor. The model can sustain longer development workflows with fewer corrections and deliver more structured outputs in tasks such as document drafting, interface design and analytical processes.

The update also includes enhancements to vision capabilities. Opus 4.7 can process higher-resolution images of up to 2,576 pixels, enabling it to analyse dense screenshots, read smaller text and interpret diagrams more effectively.

Also read: OpenAI upgrades Codex with desktop control features amid rivalry with Anthropic

Anthropic has also made changes to model behaviour and safety. The new version shows improved alignment, including stricter adherence to user instructions and increased resistance to prompt injection attempts. Additional guardrails have been introduced to limit high-risk outputs, particularly in areas such as cybersecurity.

Compared with Claude Opus 4.6, the new model delivers improvements across coding, reasoning and usability. It handles longer and more complex tasks with reduced supervision, offers better structured outputs, and demonstrates more consistent instruction following.