Startup Anthropic says its new AI model can code for hours at a time

Artificial intelligence (AI) lab Anthropic unveiled its latest top-of-the-line technology called Claude Opus 4 on Thursday, which it says can write computer code autonomously for much longer than its prior systems.

The startup, backed by Google-parent Alphabet and Amazon.com, has distinguished its work in part by building AI that excels at coding. It also announced another AI model Claude Sonnet 4, Opus's smaller and more cost-effective cousin.

Chief product officer Mike Krieger called the release a milestone in Anthropic's work to make increasingly autonomous AI. He said in an interview with Reuters that customer Rakuten had Opus 4 coding for nearly seven hours, while an Anthropic researcher set up the AI model to play 24 hours of a Pokemon game. That's up from about 45 minutes of game play for its prior model Claude 3.7 Sonnet, Anthropic told MIT Technology Review.

"For AI to really have the economic and productivity impact that I think it can have, the models do need to be able to work autonomously and work coherently for that (longer) amount of time," he said.

The news follows a flurry of other AI announcements this week, including from Google, with which Anthropic also competes.

Anthropic also said its new AI models can give near-instant answers or take longer to reason through questions, as well as do web search. And it said its Claude Code tool for software developers was now generally available after Anthropic had previewed it in February.

Reuters