Anthropic debuts Claude Sonnet 4.5 for coding and autonomous agents

Anthropic on Monday unveiled Claude Sonnet 4.5, a new large language model the company says is optimized for coding and building complex, tool-enabled AI agents. The model is available through Anthropic’s Claude API, the Claude chatbot, and integrations such as Claude Code, and Anthropic said pricing will remain at 3 per million input tokens and 15 per million output tokens.

Anthropic described Sonnet 4.5 as offering state of the art performance on several coding benchmarks and better reliability for long running tasks. The company reported a SWEBench Verified success rate of 77.2 percent for Sonnet 4.5, up from 74.5 percent for its prior Opus 4.1 model, and provided additional scores on task suites the company uses to measure real world engineering ability. Anthropic also said the model sustained autonomous operation for more than 30 hours in internal trials, during which it wrote thousands of lines of code to build a chat application and carried out ancillary tasks such as configuring databases, purchasing domain names, and performing a SOC 2 audit to verify security controls. Anthropic AI researcher David Hershey said, "It is hard to capture Claude Sonnet 4.5's performance...