Claude 4.1 tops benchmarks

Anthropic's Claude Opus 4.1 has been released, demonstrating advancements in coding, reasoning, and agentic tasks. The model achieves a 74.5% score on the SWE-bench Verified benchmark, a notable improvement in coding capabilities. This upgrade shows gains in multi-file code refactoring, debugging, research, and data analysis. Rakuten Group and GitHub have already integrated Opus 4.1, reporting positive results in debugging and code refactoring.

Opus 4.1 is available to paid Claude users, through Anthropic's API, and on platforms like Amazon Bedrock and Google Cloud's Vertex AI. Pricing remains consistent with the prior Opus 4 version, at $15 per million input tokens and $75 per million output tokens. Anthropic plans further model improvements in the coming weeks.

While Claude 4 models show great promise, nearly half of Anthropic's API revenue relies on only two customers.