METR published an assessment on November 5 estimating that Anthropic's internal AECI benchmark trend accelerated 1.9 times after October 2025, with a 95% confidence interval of 1.5–2.8 times. The finding concerns selected capability benchmarks, rather than every kind of real-world task.
The researchers judge that AI likely explains much of the current acceleration, but describe that attribution as uncertain. Anthropic also redacted much of the evidence METR considered most informative. Readers therefore cannot reproduce the full assessment from its public report alone.
A faster benchmark trend and a causal explanation are different propositions. Establishing the first does not isolate the effects of tools, staffing, research choices or other changes in the development process. A confidence interval for a measured trend also does not capture every uncertainty about the interpretation of that trend.
The report is significant because it makes the pace of AI-assisted research a subject of external evaluation. Its disclosure limits are equally important: public conclusions should remain narrower than claims that AI can already replace a research organization or that faster progress must continue. Further assessments with comparable methods and more reviewable evidence would help separate those questions.