METR

What METR found about the pace of AI research at Anthropic

An outside assessment found Anthropic’s models improving about twice as fast on defined tasks. Much of the evidence behind the risk findings remains private.