Anthropic unveils the latest Claudes with claim to AI coding crown
Add Axios as your preferred source to
see more of our stories on Google.
Anthropic chief product officer Mike Krieger at Thursday's San Francisco event. Photo: Ina Fried/Axios
Anthropic debuted the first models in its latest Claude 4 series Thursday — including one, Claude 4 Opus, that it says is the world's best at coding.
Why it matters: Competition is hot between Anthropic, Google and OpenAI for the "best frontier model" crown as questions persist about the companies' ability to push current AI techniques to new heights.
Driving the news: At the high end, Anthropic announced Claude 4 Opus, its "powerful, large model for complex challenges," which it says can perform thousands of steps over hours of work without losing focus.
- Claude Sonnet 4 — Anthropic's "smart, efficient model for everyday use" — is designed to replace Claude Sonnet 3.7 with improved coding abilities and better adherence to instructions.
- Both are hybrid models, meaning they can answer users straightaway or take additional compute time to perform reasoning steps.
- In addition to the new models, Anthropic announced a series of other new capabilities for developers who connect to its models, including the ability to combine reasoning with the use of the Web and other tools.
What they're saying: "AI agents powered by Opus 4 and Sonnet 4 can analyze thousands of data sources, execute long-running tasks, write human-quality content, and perform complex actions," Anthropic said in a statement.
Between the lines: Anthropic said it had to institute new safety controls over Claude 4 because of its power.
- "The model is so strong that we had to turn on additional safety mitigations according to Anthropic's responsible scaling policy," researcher Jan Leike said in a post on X.
Anthropic is also changing its models' reasoning mechanics — it will now aim to show summaries of the models' thought processes rather than trying to document each step.
- "We've made this update to preserve visibility for users while better securing our models," Anthropic said in a statement.
The big picture: The announcements, made at Anthropic's first-ever developer conference, come after a busy week in AI that saw Microsoft announce a new coding agent and a partnership to host Elon Musk's Grok, Google expand its AI-powered search efforts and OpenAI announce a $6.5 billion deal to buy io, Jony Ive's secretive AI hardware startup.
