Today, we publish Claude Opus 4.1, an upgrade to Claude Opus 4 on agency tasks, real coding and reasoning. We plan to publish considerably greater improvements in our models in the coming weeks.
OPUS 4.1 is now available for Claude paid users and in the Claude code. It is also located on our API, our Rocky substratum from Amazon and the top of Google Cloud. The price is the same as opus 4.
Close work 4.1
OPUS 4.1 advances our cutting-edge coding performance at 74.5% on verified Swe-Bench. It also improves in -depth research and data analysis skills of Claude, in particular around the monitoring of details and agent research.
Girub Note that Claude Opus 4.1 improves in most capacities compared to Opus 4, with particularly notable performance gains in the reckorization of multi-fichiers code. Rakuten group note that OPUS 4.1 excels in the training of exact corrections in large code bases without making unnecessary adjustments or the introduction of bugs, their team preferring this precision for daily debugging tasks. Windsurfing OPUS 4.1 reports deliver an improvement in standard deviation from opus 4 on their junior developer reference, showing roughly the same performance jump as the Sonnet 3.7 jump to Sonnet 4.
To start
We recommend the upgrade of opus 4 to opus 4.1 for all uses. If you are a developer, just use claude-opus-4-1-20250805 via the API. You can also explore our system card, our model page, our price page and our documents to find out more.
As always, your comments help us improve, especially since we continue to publish new and more capable models.
