Anthropic launches Claude Sonnet 4.5, its most powerful model for coding

Anthropic has unveiled Claude Sonnet 4.5, its latest frontier AI model, which the company says delivers state-of-the-art performance in coding and reliability strong enough to power “production-ready” applications—not just prototypes. Available now through the Claude API and chatbot, pricing remains unchanged from Claude Sonnet 4 at $3 per million input tokens and $15 per million output tokens.

Over the past year, Anthropic has carved out a reputation as a go-to for enterprises and developers tackling software engineering tasks. Its models are reportedly used internally at Apple and Meta, while powering developer tools like Cursor, Windsurf, and Replit. But the competitive landscape has shifted: OpenAI’s GPT-5 has recently bested Claude models on several coding benchmarks.

According to a report by Tech Crunch, Anthropic insists Claude Sonnet 4.5 reclaims the edge. The model scored strongly on industry tests like SWE-Bench Verified, though Anthropic researcher David Hershey argues its capabilities extend beyond benchmarks. As per the report, he describes the model autonomously coding for 30 straight hours in enterprise trials—standing up database services, buying domain names, and even running a SOC 2 audit to validate security. Cursor CEO Michael Truell, in a statement to Tech Crunch, called it “state-of-the-art on longer horizon tasks,” while Windsurf’s Jeff Wang said it signalled a “new generation of coding models.”

Beyond raw coding chops, Anthropic claims Claude Sonnet 4.5 is its most aligned model to date, with lower rates of sycophancy and deception, and improved defences against prompt injection attacks.