Outside study finds Anthropic AI improving twice as fast

Anthropic is improving its AI models about twice as fast as it did a year ago, according to an independent assessment released Thursday by the nonprofit METR. The finding comes weeks after CEO Dario Amodei urged the industry to slow its pace. The report is one of the most detailed outside looks yet at whether AI is speeding up its own development. It arrives as Anthropic prepares for an initial public offering expected later this month, the Associated Press reported.abcnews.com

What METR found

METR staff ran a "high-access evaluation" at Anthropic from Aug. 12 to Oct. 29. METR's report concludes that as of October, the company is "likely improving the capabilities of their models on well-specified tasks at around twice the rate" it did between March 2024 and October 2025. The group measured progress with Anthropic's internal benchmark index and found scores have grown about 1.9 times faster since October 2025, with a 95% confidence interval of 1.5 to 2.8 times. In other words, Anthropic has made roughly two years of its earlier progress in the past year.metr.org

METR said it was "confident" that faster growth in training compute did not cause the speedup. It said most of the current acceleration likely comes from Anthropic's heavy use of AI in its own research. Reuters reported that METR put AI's share of the acceleration at perhaps 80%, while calling that figure uncertain. Researchers also found no convincing evidence of major bottlenecks that could slow progress over the next year.reuters.com

The report came with caveats. METR said the true rate could be closer to 1.3 or 3 times the historical pace. The speedup might apply only to tasks that can be graded automatically. Anthropic's redactions also limited how much detail the group could publish, AP reported.abcnews.com

Disagreement over Anthropic's own risk report

In a separate review, METR said Anthropic's August risk report "understates the level of acceleration that was occurring." It said Anthropic's internal "2x Acceleration Threshold" had plausibly been crossed by July 15. METR still agreed that the direct catastrophic risks from those models were low. It added that it "cannot rule out a substantial further acceleration" over the next six months.metr.org

Amodei's call to slow down

In a Sept. 12 essay, Amodei urged the AI world to "slow the pace at which we improve the capabilities of AI models." He warned that recursive self-improvement "could outrun our ability to understand and control these systems." In the same essay he named METR as a suitable outside evaluator, according to AP. The company has also published its own research showing that AI is already accelerating its development work.anthropic.com+1

Anthropic did not immediately respond to requests for comment. METR said AI cannot yet make "highly significant and fully autonomous contributions" to model training, but noted that Anthropic researchers described "signs of life" suggesting that could change soon.