Anthropic: Claude Sonnet 4

Claude Sonnet 4 is a hybrid reasoning model that combines near-instant standard responses with optional extended thinking for deeper problem-solving, giving it the flexibility to handle both quick queries and complex multi-step tasks in a single session. A key design focus was enhanced steerability, meaning the model responds more precisely to detailed instructions and complex chains of intent, making it notably better at autonomous multi-feature application development than its predecessor. Its strengths in coding and reasoning are backed by a 72.7% score on SWE-bench Verified, a benchmark that measures real-world software engineering performance on GitHub issues.

Released in May 2025 alongside Claude Opus 4, Sonnet 4 builds directly on Sonnet 3.7 as a significant step forward in capability and reliability. The model was adopted by GitHub to power the new coding agent in GitHub Copilot, and third-party testers reported navigation errors dropping from 20% to near zero in agent-driven workflows, reflecting its improved reliability for autonomous tasks. It introduced one-hour prompt caching as a new API capability, reducing costs for repeated contexts. Optimized for practical everyday use while maintaining computational efficiency, Sonnet 4 balances advanced reasoning with responsiveness across a wide range of internal and external applications.