Anthropic launchs Claude Sonnet 4.5, calls it the "best coding model in the world"

Anthropic rolled out Claude Sonnet 4.5 on Monday, its newest AI model that the company says delivers stronger performance in coding, computer use, and practical business applications.

The startup, which has backing from Amazon and carries a valuation of $183 billion, is making Claude Sonnet 4.5 accessible to everyone who uses its platform.

According to the company, the model shows particular strength in specialized areas including cybersecurity, finance, and research work.

Anthropic is calling Claude Sonnet 4.5 the “best coding model in the world” based on industry testing standards such as SWE-bench Verified, which evaluates how well AI systems can handle software coding tasks.

Jared Kaplan, who co-founded Anthropic and serves as chief science officer, spoke with CNBC about the release.

“People are just noticing with this model, because it’s just smarter and more of a colleague, that it’s kind of fun to work with it when encountering problems and fixing them,” Kaplan explained.

The company says the model produces better quality code, does a better job spotting ways to improve existing code, and follows directions more consistently than previous versions.

Continuing fast-paced innovation

This launch follows Anthropic’s August release of Claude Opus 4.1 and May release of Claude Sonnet 4. It highlights the rapid speed at which AI companies are developing new technology.

A group of researchers who previously worked at OpenAI started Anthropic in 2021. Since then, the two have emerged as strong AI rivals.

OpenAI launched the generative AI revolution when it released ChatGPT in 2022. That startup, now valued at $500 billion, put out its latest version called GPT-5 last month. However, that launch faced problems when some users reported losing access to earlier models.

Mike Krieger, who is chief product officer at Anthropic, said Claude Sonnet 4.5 will become the standard option for users. He noted the company suggests using this model for “basically every use case.”
Still, users have choices. People with paid subscriptions can continue using Opus, and those with particular workflows can pick an older Sonnet version if they need more time before switching, Krieger said.

Krieger explained that despite being a smaller model than Claude Opus 4.1, Claude Sonnet 4.5 delivers superior results in almost all areas. He said both the company and its customers find it “very useful for real, actual work”.

The new model can operate on its own for thirty hours while staying focused on complicated, multi-step tasks throughout that time. By comparison, Claude Opus 4, which came out in May, could only run independently for seven hours.

Safety improvements take center stage

Anthropic also improved the model’s safety through extensive training. The company says it reduced problematic behaviors, including deception, power seeking, and sycophancy, which happens when a model simply tells users what they want to hear.

Claude Sonnet 4.5 also has better defenses against prompt injection attacks, where someone tricks a model into doing something harmful, like revealing confidential information.

“This is the biggest jump in safety that I think we’ve seen in the last probably year, year and a half,” Kaplan said.

More releases are coming from Anthropic soon. Kaplan mentioned better models are in the works, including “very likely Opus.”

“No promises,” he said. “But I think that we’ll probably have one or two more releases before the end of the year.”

If you're reading this, you’re already ahead. Stay there with our newsletter.