Anthropic has launched Claude Sonnet 4.5, its most advanced AI model yet for software development. The company claims this new iteration can create fully operational, production-grade applications with greater reliability than previous versions.
Available through the Claude API and chatbot, Claude Sonnet 4.5 is priced identically to its predecessor: $3 per million input tokens (about 750,000 words) and $15 per million output tokens. This pricing keeps Anthropic competitive in the AI coding market.
Over the past year, Anthropic’s AI has gained popularity among developers and enterprises, including Apple and Meta, for its strong software engineering capabilities. Platforms like Cursor, Windsurf, and Replit also rely on its API. However, the recent launch of OpenAI’s GPT-5 has intensified competition by outperforming Claude models across various coding tests.
Anthropic highlights Claude Sonnet 4.5’s top-tier performance on coding benchmarks such as SWE-Bench Verified but notes that benchmarks alone do not fully capture its abilities.
Researcher David Hershey observed the model autonomously coding for up to 30 hours in enterprise trials, building applications, setting up databases, buying domain names, and completing SOC 2 security audits.
Industry leaders praised the model’s capabilities. Cursor CEO Michael Truell called it “state-of-the-art coding performance” for long-term projects, while Windsurf CEO Jeff Wang described it as “a new generation of coding models.”
Anthropic also says Claude Sonnet 4.5 is its most aligned AI yet, with fewer issues like sycophancy and dishonesty, and improved defence against prompt injection attacks. Alongside the model, Anthropic released the Claude Agent SDK, enabling developers to build customised AI agents.
Additionally, the company launched a limited preview called “Imagine with Claude” for Max subscribers, allowing the AI to generate software dynamically in response to user prompts without pre-written code.