August 6, 2011 - Anthropic, Inc. today (August 6) announced the launch of the Claude Opus 4.1 model, compared to the Claude 4 series model introduced in May of this year.Major improvements in coding, reasoning, and executing instructions.
Anthropic says that Claude Opus 4.1 improved to 74.51 TP3T on SWE-bench Verified (which is used to assess software engineering accuracy), compared to 62.31 TP3T for Claude Sonnet 3.7 and 72.51 TP3T for Claude Opus 4. Specifically, the updated model performs better in terms of "in-depth research and data analysis skills, especially in detail tracking and agent search".
Compared to Opus 4, Opus 4.1 improves on most of the features, especially on the multi-fileCodeThe reconfiguration aspect is particularly well represented.
Rakuten Group found that Opus 4.1 was able to pinpoint and fix bugs in a large code base without making unnecessary adjustments or introducing new bugs.
Windsurf reports that Opus 4.1 improved by one unit of standard deviation over Opus 4 performance in its Junior Developer Benchmarks, a performance leap comparable to the jump from Sonnet 3.7 to Sonnet 4.
The latest model is available to Claude customers starting today and can be used through Claude Code, Anthropic's API, Amazon Bedrock and Google Cloud's Vertex AI.
Meanwhile, Anthropic said on social media that it plans to release "significant improvements to our models" in the coming weeks, so expect more upgrades to the Claude family of models, and OpenAI is also expected to make a new announcement this week.