Anthropic's models now improve twice as fast — mostly thanks to AI

METR, the nonprofit AI evaluator, published an independent assessment on November 5 finding Anthropic is likely improving its models on well-specified tasks at about twice the rate it did between March 2024 and October 2025 — roughly two years of historical progress in a single year, with benchmark scores growing 1.9 times faster since October 2025.

METR's subjective estimate is that perhaps 80% of the current acceleration comes from Anthropic's own AI doing the research — staff self-reports of speedup reached 12x in October — and it is confident the jump was not caused by faster growth in training compute. It found no clear bottleneck likely to slow things in the next 12 months.

The twist: Anthropic redacted a majority of the evidence METR found most informative, so the public has to take METR's word for it — and METR says it is plausible Anthropic already crossed its own 2x acceleration early-warning threshold, a claim the company's August risk report understates.