Anthropic announced on Monday the launch of Claude Sonnet 4.5, a new frontier model the company claims delivers state-of-the-art performance on coding benchmarks. The company stated that the model can build “production-ready” applications, representing an improvement in reliability over previous AI models. Claude Sonnet 4.5 is accessible through the Claude API and the Claude chatbot, with pricing for developers remaining the same as for Claude Sonnet 4: $3 per million input tokens and $15 per million output tokens.
Anthropic’s AI models are used by developers and enterprises, with reports that Apple and Meta utilize them internally. The company also sells API access to AI coding applications such as Cursor, Windsurf, and Replit. This release follows recent reports that OpenAI’s GPT-5 has challenged Anthropic’s models by outperforming them on several coding benchmarks.
According to Anthropic, Claude Sonnet 4.5 achieves industry-leading performance on benchmarks like SWE-Bench Verified. Anthropic AI researcher David Hershey stated that benchmark results alone do not capture the model’s full capabilities. Hershey reported observing the model code autonomously for up to 30 hours during early trials with enterprise customers. In these sessions, the AI built an application, set up database services, purchased domain names, and performed a SOC 2 audit. Michael Truell, CEO of Cursor, described Claude Sonnet 4.5 as having “state-of-the-art coding performance, specifically on longer horizon tasks.” Jeff Wang, CEO of Windsurf, called the model a “new generation of coding models.”
Anthropic also claims Claude Sonnet 4.5 is its most aligned frontier AI model yet, with lower rates of sycophancy and deception than its predecessors. The company said it has improved the model’s resistance to prompt-injection attacks.
