METR estimates Anthropic's task capabilities are improving faster than before

In an assessment published on 5 November, METR estimated that Anthropic's progress on well-defined task capabilities is now running at about 1.9 times its previous pace.

The estimate comes with significant uncertainty: the 95 per cent confidence interval ranges from 1.5 times to 2.8 times the earlier rate of progress.

Much of the underlying evidence behind the assessment has not been made public, which limits how far outside observers can check the figures.

Why it matters: the estimate bears on how quickly AI systems can improve at defined technical tasks. It does not prove that autonomous recursive self-improvement is under way, and the limited public evidence makes independent assessment difficult too.