Anthropic Unveils Claude Sonnet 4.5: The World's Best Coding Model
Anthropic has unveiled Claude Sonnet 4.5, the latest iteration of its mainstream large language model, marking a significant advancement for software developers, particularly in code generation and reasoning. The model was launched on September 29-30, 2025, with availability commencing on the same date. Anthropic has hailed Sonnet 4.5 as the 'best coding model in the world'.
Sonnet 4.5 has demonstrated impressive performance on various benchmarks. On the OSWorld benchmark, it achieved a success rate of 61.4%, leading the charts. It also scored an impressive 77.2% on SWE-Bench Verified, outperforming its predecessors and competitors. The model's capabilities have been further enhanced with new features like virtual machines, memory, and multi-agent support. Additionally, Sonnet 4.5 can now run autonomously for 30 hours, a significant improvement from the seven hours of Opus 4.1.
Anthropic is also expanding its offerings with the launch of the Claude Agent SDK. This will enable developers to build agents using the same infrastructure as Claude Code. Furthermore, Anthropic's coding agent, Claude Code, will also gain access to Sonnet 4.5 and receive new features. The company's experiment, 'Imagine with Claude', will use Sonnet 4.5 to generate software and user interfaces on the fly.
Claude Sonnet 4.5's release brings substantial improvements to Anthropic's large language model suite, promising enhanced capabilities for software developers. With its impressive benchmark scores, extended autonomous running time, and new features, Sonnet 4.5 is set to revolutionize coding and software generation.
