METR estimates Anthropic is improving its models twice as fast
An outside assessment with access to Anthropic’s internal data puts the acceleration of its model development at about 2x since October 2025, mostly from the use of AI in research. It came three days after Anthropic’s amended filing said the company may soon meet its own threshold for AI-driven acceleration.
Source published · In the Weekly Review of 8 November 2026
Filed under AI safety, AI industry
On 2 November Anthropic added a passage to the risk factors of its amended listing document. Internal estimates and expectations about future models, it said, point towards the threshold in its Responsible Scaling Policy for AI that significantly speeds up AI development, although no current model is designated above it. The company said it has failed before to notice a significant acceleration until after it happened, and that it is considering policy changes to keep humans in control.
METR, a non-profit evaluator, published its own assessment on 5 November after spending from 12 August to 29 October inside Anthropic’s research data. It concluded that progress on well-specified tasks is running at about twice the rate of March 2024 to October 2025, and would not be surprised by a true figure between 1.3x and 3x. Scores on Anthropic’s version of the Epoch Capabilities Index grew at 1.9 times the earlier rate, with a 95% confidence interval of 1.5 to 2.8, and METR ruled out faster growth in training compute as the cause.
The evidence is partly hidden. Anthropic redacted most of what METR considered most informative, and METR noted that the index favours verifiable tasks and may overstate progress.