AI startup Anthropic has announced the release of Claude Sonnet 4.5, describing it as its strongest model yet for software development. The company says the model is capable of generating not just prototypes but fully production-ready applications.

Claude Sonnet 4.5 is available starting today through the Claude chatbot and API, priced at $3 per million input tokens and $15 per million output tokens, unchanged from the previous version.

According to Anthropic, the model leads on multiple coding benchmarks including SWE-Bench Verified, although researchers noted that performance in real-world software tasks often goes beyond benchmarks. In internal trials, Sonnet 4.5 was able to code autonomously for up to 30 hours, setting up database services, purchasing domain names, and even conducting a SOC 2 security audit without human intervention.

The launch also introduces the Claude Agent SDK, a toolkit designed to help developers create intelligent AI agents using the same infrastructure that powers Claude Code. In addition, Anthropic is offering a research preview feature called “Imagine with Claude” for Max subscribers, which enables real-time software generation from natural language prompts.

Anthropic says Sonnet 4.5 includes improvements in alignment, safety, and robustness, addressing issues like sycophantic behavior and prompt injection attacks. These updates are aimed at making the model more reliable for enterprise and developer use.

The release comes just two months after Anthropic unveiled Claude Opus 4.1, highlighting the rapid pace of AI development in 2025. The company faces stiff competition from rivals including OpenAI, whose GPT-5 has gained ground in several coding benchmarks.

Anthropic’s models are already used by major firms such as Apple, Meta, and development platforms including Cursor, Windsurf, and Replit. With Sonnet 4.5, the company is positioning itself as a leader in bridging the gap between prototype code and production-ready applications.