Anthropic introduced an updated version of its flagship AI model Claude Opus 4.1. It became available to users of paid plans Claude, Claude Code, as well as through API, Amazon Bedrock, and Google Cloud Vertex AI. The subscription cost remained at the level of the previous version Opus 4. The model received improvements in programming, including better performance during code refactoring, analysis of large data volumes, and execution of complex multi-step tasks.
Claude Opus 4.1 demonstrated a record result in the SWE-bench Verified test, reaching 74.5 percent. This is approximately two percent higher than the Opus 4 score and five percent ahead of the o3 model from OpenAI. The test evaluates the AI’s ability to find and fix real errors in open-access code. In addition to programming, Claude Opus 4.1 became more accurate in analysis and research tasks, better tracking details and conducting agent-style searches more efficiently.
Developers noted a noticeable increase in Claude Opus 4.1’s performance in junior specialist tasks, and companies GitHub and Rakuten Group noted the model’s accuracy when working with large codebases. Claude Opus 4.1 can process up to 64,000 tokens at a time, allowing it to solve more complex tasks without losing quality.
Anthropic implemented the highest level of protection for Claude Opus 4.1 according to its own ASL-3 security system. This includes additional measures against model theft and misuse. The company recommends all users switch to Opus 4.1 and reports plans to significantly improve Claude’s capabilities in the near future.
