Claude Sonnet 4.5 just made its entrance, and its most advanced AI model yet for software engineering.
Developers and enterprises have been turning to Anthropic’s models in droves, eager for that extra edge in building applications. Pricing for the new model mirrors its predecessor, with developers looking at three dollars per million input tokens and fifteen dollars per million output tokens.
Apple and Meta have reportedly leaned into Anthropic tech for internal projects. Not only that, but names like Cursor, Windsurf, and Replit are relying on Anthropic’s models for building coding tools and integrations.
Raising the Bar for AI Coding
Anthropic is confident Claude Sonnet 4.5 raises the bar on current coding benchmarks and can actually deliver production grade applications, not just ambitious prototypes. It touts performance on tests like SWE Bench Verified as among the best in the industry.
David Hershey, an AI researcher at Anthropic, says pure benchmarks only tell part of the story. He shared that in live trials, “We’ve seen Claude Sonnet 4.5 take on entire projects for as long as 30 hours, wiring up databases, purchasing domain names, and even running security audits without missing a beat.”
On the client side, companies working closely with Anthropic are equally enthusiastic. As Cursor CEO Micheal Truell puts it, Sonnet 4.5 “offers state of the art coding performance, especially on longer horizon tasks.”
The new release also puts a big emphasis on trust and alignment. Anthropic claims this version is less likely to agree blindly or provide deceptive responses, a common issue with earlier AI models. The model has also been updated with extra guards to make it less vulnerable to prompt injection attacks that can trick AI into misbehaving.
Along with the new model, Anthropic rolled out the Claude Agent SDK, an infrastructure that powers Claude Code. With this, developers will be able to create custom AI agents for their own use cases.
Anthropic is experimenting with a research preview called “Imagine with Claude,” which is available to Max subscribers. This feature lets users see the AI generate software instantly in response to prompts, promising a live coding experience and no canned code.
AI competition is fierce right now, with OpenAI’s GPT 5 now in the race and Anthropic pushing out major model upgrades every few months. OpenAI’s GPT 5 now in the race and Anthropic pushing out major model upgrades every few months. Claude Sonnet 4.5 comes less than sixty days after the last major release, underlining how quick the cycle has become.
Holding onto a lead in this shifting landscape seems more challenging than ever, but Anthropic appears determined to stay in the conversation.
