Anthropic drops Claude Opus 4.1 with a massive coding boost

Claude Opus 4.1 just dropped, and Anthropic is taking direct aim at messy, real-world coding.

While this looks like a minor point update to Claude Opus 4, it brings targeted upgrades to agentic tasks, in-depth research, and software engineering. It is already live for paid Claude users, inside Claude Code, via the API, and on both Amazon Bedrock and Google Cloud's Vertex AI. Best of all, pricing stays identical to Opus 4.

Precision surgery on huge codebases

The headline number is a 74.5% score on SWE-bench Verified.

Notably, Anthropic hit this mark without using extended thinking. The setup was remarkably lean: just a bash tool and a string-replacement file editor, ditching the separate planning tool used back in Sonnet 3.7. The team also evaluated across the full 500 problems, pointing out that OpenAI models are evaluated on a 477-problem subset.

Early feedback from developer platforms shows where that raw score actually matters.

GitHub noted broad gains over Opus 4, highlighting massive improvements in multi-file code refactoring.

Rakuten Group reported that Opus 4.1 excels at pinpointing exact fixes inside massive codebases without making unnecessary adjustments or spawning new bugs. Their engineers now prefer it for everyday debugging.

AI code editor Windsurf measured a full standard deviation jump on its junior developer benchmark, calling it a leap roughly comparable to the shift from Sonnet 3.7 to Sonnet 4.

Beyond coding, Anthropic also tuned the model for agentic search and detail tracking during deep analytical tasks.

Why it matters

We think this release solves the single biggest complaint engineers have with coding agents: collateral damage.

Most models can generate isolated functions, but they often butcher existing architecture or introduce regression bugs when touching legacy code. The fact that partners like Rakuten emphasize surgical precision over flashy generation is a massive signal that agentic coding is maturing into something safe for production.

If you build software or spend hours refactoring multi-repo setups, it is well worth switching your API key to Opus 4.1 today. The upgrade is essentially free, and the reduced debugging overhead will pay for itself immediately.

Strategically, Anthropic is cementing Claude's role as the undisputed default engine for modern developer tooling. But the most intriguing detail was Anthropic's parting note: they explicitly teased substantially larger model upgrades arriving in the coming weeks. Opus 4.1 is just the warm-up.