Claude Opus 4.7 is now generally available, offering a leap over its 4.6 predecessor in software engineering. Early testers say the model can take on the toughest coding assignments without close supervision, handling long‑running, multi‑step jobs with strict instruction following. Enhanced vision supports higher‑resolution images, producing sharper interfaces, slides and documents. It also verifies its own outputs before returning results, reducing manual review.

Following Project Glasswing, Anthropic limited the more powerful Claude Mythos Preview and used Opus 4.7 as the first model to trial new cyber safeguards. The system automatically blocks high‑risk security requests, while a Cyber Verification Program invites professionals to test legitimate penetration‑testing use cases. Opus 4.7 ships on Claude’s suite, Amazon Bedrock, Google Vertex AI and Microsoft Foundry at $5 per million input tokens.

Internal benchmarks show Opus 4.7 lifting coding resolution by 13% on a 93‑task suite, solving four problems that stumped both Opus 4.6 and Sonnet 4.6. Multi‑step efficiency rose to a 0.715 score, with finance module accuracy climbing to 0.813. Developers report faster, more reliable tool calls, making the model feel like a true coding teammate.