Anthropic today announced the availability of Claude Sonnet 4.5, its latest large language model. Anthropic claims that Sonnet 4.5 is the best coding model in the world and is particularly well-suited for building complex agents.
On the SWE-bench Verified coding benchmark, which evaluates real-world software development abilities, Claude Sonnet 4.5 scored a state-of-the-art 82%. During internal testing, Anthropic found that Sonnet 4.5 was able to run continuously in agent mode for more than 30 hours. On the OSWorld benchmark, which tests AI models on real-world computer tasks, Claude Sonnet 4.5 scored 61.4%, a massive leap from the 42.2% achieved by Sonnet 4.
In Terminal-Bench, Sonnet 4.5 scored 50%, outperforming OpenAI’s GPT-5 and Google’s Gemini 2.5 Pro. However, in general AI benchmarks such as GPQA Diamond, MMMLU, and MMMU, the new Sonnet 4.5 lags behind GPT-5 and Gemini 2.5 Pro. Given the popularity of Claude models among developers, these benchmarks suggest that Anthropic is focusing on improving its models for coding-related tasks.
Claude Sonnet 4.5 is available as claude-sonnet-4-5 via the Claude API. Pricing remains the same as Claude Sonnet 4: $3 per million input tokens and $15 per million output tokens. Anthropic recommends everyone upgrade to Claude Sonnet 4.5 for all uses since it provides much improved performance for the same price.
Microsoft’s GitHub today announced the availability of Claude Sonnet 4.5 in GitHub Copilot for Copilot Pro, Pro+, Business, and Enterprise users. The GitHub team stated that this new model delivers major improvements in tool orchestration, context editing, and domain-specific capabilities. Additionally, Copilot Free, Pro, and Pro+ users can bring their own key in Visual Studio Code to access this latest model from Anthropic.
Finally, the Sonnet 4.5 model is also available for developers via Amazon Bedrock, a fully managed service that offers foundation models via APIs.