The short version
● Headline — roughly 2x faster capabilities progress
● On the index — 1.9x, 95% CI 1.5 to 2.8x
● Share from AI — best guess around 80%
● The catch — Anthropic redacted the strongest evidence
METR published an independent assessment on 5 November concluding that Anthropic is likely improving its models on well-specified tasks at around twice the rate it managed from March 2024 to October 2025. It adds that it would not be surprised if the real figure were closer to 1.3x or 3x.
The numbers behind it: Anthropic's internal benchmark index, AECI, shows 1.9x acceleration with a 95% confidence interval of 1.5 to 2.8x, with the trend rising from about +15 points a year to about +29, and a breakpoint on 7 October 2025. METR's best guess is that a majority, perhaps around 80%, of the October 2026 acceleration is due to AI, and it is confident the gain did not come from scaling training compute. Widen the sample to all frontier developers and the estimate drops to about 1.4x. Fieldwork ran from 12 August to 29 October with roughly 2.5 full-time-equivalent staff, about 15 interviews, and internal benchmark, headcount, code and inference-spend data.
METR is unusually direct about what it could not do. Anthropic redacted much of the evidence METR found most informative, so the public cannot independently verify the key conclusions. Robustness checks produced estimates from 1.1x to 2.9x depending mainly on the link function assumed for benchmark scores. Staff self-reported uplift reached about 12x for the most capable internal model, which METR calls only very loose evidence. It found no convincing bottleneck in algorithms, compute or data over the next 12 months, while noting its own view on that changed several times.
More from 6 November
Full dated index: every AI release in November 2026.
Sources