Anthropic launches Claude Opus 4.7 with sharper coding, vision, and safety controls

What’s new

Anthropic has made Claude Opus 4.7 generally available, pitching it as a focused upgrade over Opus 4.6 with particular gains in advanced software engineering, multi-step tasks, and higher-resolution vision. It has been reported that users can now hand off their hardest coding work—the stuff that used to need constant supervision—with greater confidence. The company says the model pays closer attention to instructions, reasons about verification steps, and catches logical faults during planning. Relief for engineers? Quite possibly.

Performance and early feedback

Anthropic claims Opus 4.7 produces cleaner interfaces, slides, and docs, and that it shows measurable gains on internal benchmarks. It has been reported that on a 93-task coding benchmark Opus 4.7 lifted resolution by about 13% over Opus 4.6, including solving tasks neither prior Opus nor Sonnet models could complete. Anthropic shared testimonials from early-access testers praising faster median latency, strict instruction-following, and stronger handling of long-running async workflows—critical for CI/CD and automation at scale.

Safety posture and Project Glasswing

This release also serves as a test bed for Anthropic’s staged approach to security. Following last week’s Project Glasswing announcement, Anthropic says Opus 4.7 was deliberately trained to be less capable in cyber offense than its Mythos Preview line, and it ships with automated safeguards to detect and block prohibited or high-risk cybersecurity queries. It has been reported that legitimate security researchers can apply to a new Cyber Verification Program to gain vetted access for red-teaming and vulnerability research.

Availability and pricing

Opus 4.7 is available across Claude products and via the API, and it is on Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Anthropic says pricing is unchanged from Opus 4.6: $5 per million input tokens and $25 per million output tokens, and developers can call claude-opus-4-7 through the Claude API. The rollout underscores an industry trend: capabilities raced forward, but companies are increasingly staging releases and baking in controls. Smart move, or cautious posture? Time (and usage) will tell.

Sources: anthropic