- Anthropic released Claude Opus 4.7 to general availability on May 22, 2026.
- Opus 4.7 improves over Opus 4.6 on the hardest software-engineering tasks, complex long-running tasks, and self-verification of outputs.
- Vision is substantially better — the model can see images in greater resolution.
- Cyber-safeguards automatically detect and block requests that indicate prohibited or high-risk cybersecurity uses.
What Happened
Anthropic released Claude Opus 4.7 to general availability on Thursday. The release improves on Opus 4.6 in advanced software engineering with particular gains on the most difficult tasks. The company says users can now hand off their hardest coding work — the kind that previously needed close supervision — to Opus 4.7 with confidence.
Why It Matters
Opus 4.7 lands during one of the most active product weeks in Anthropic’s history: Code with Claude in London May 19-20 showed half the developer audience had shipped Claude-written pull requests in the past week; KPMG announced a 276,000-employee global Claude rollout the same day; PwC expanded its Claude deployment with a 30,000-professional training program; and the company disclosed it expects its first profitable quarter at roughly $10.9 billion in Q2 revenue.
The cyber-safeguard architecture is the most novel piece of the release. Last week Anthropic announced Project Glasswing, highlighting the risks and benefits of AI models for cybersecurity. The company committed to keeping Claude Mythos Preview’s release limited and testing new cyber safeguards on less capable models first. Opus 4.7 is the first such model.
Technical Details
Per the announcement, Opus 4.7 handles complex, long-running tasks with rigor and consistency, pays precise attention to instructions, and devises ways to verify its own outputs before reporting back. The model also has substantially better vision and can see images in greater resolution. It is more tasteful and creative when completing professional tasks, producing higher-quality interfaces, slides, and documents. Anthropic stated Opus 4.7 is less broadly capable than the most powerful Claude Mythos Preview but shows better results than Opus 4.6 across a range of benchmarks.
For cyber capabilities specifically, Anthropic notes that Opus 4.7’s capabilities are not as advanced as those of Mythos Preview — during its training the company experimented with efforts to differentially reduce these capabilities. The cyber-safeguards automatically detect and block requests that indicate prohibited or high-risk cybersecurity uses. Security professionals who wish to use Opus 4.7 for legitimate cybersecurity purposes such as vulnerability research and penetration testing have a separate access pathway.
Who’s Affected
Anthropic enterprise customers — including KPMG (276K workforce), PwC (hundreds of thousands of professionals), and the long list of Claude Code deployments — gain a more capable base model. Developers using Claude Code see the same capability tier as Opus 4.7 within their existing tooling. Security professionals see a model that has differentially-reduced cyber capabilities by design, with an opt-in pathway for legitimate research. Competing model providers — OpenAI’s GPT-5.5, Google’s Gemini 3 Pro — face a new comparison point on coding benchmarks. The Cursor Composer 2.5 release on May 17, which matched Opus 4.7 on SWE-Bench Multilingual, now faces a higher target.
What’s Next
Opus 4.7 is now generally available through the Claude API, Claude.ai, and Anthropic enterprise partners. The cyber-safeguard architecture is the first deployment of Anthropic’s Project Glasswing framework; what the company learns from real-world safeguard deployment will inform the eventual broad release of Mythos-class models. Industry watchers should expect comparison benchmarks against GPT-5.5 and Gemini 3 Pro in coming days from independent evaluators.
