METR estimates Anthropic benchmark gains accelerated 1.9 times since October

METR published an independent assessment on November 5, 2026, estimating Anthropic’s capabilities on well-specified benchmark tasks improved about 1.9 times faster since October 2025 than between March 2024 and October 2025. AECI scores rose from about 15 points per year to about 29 after a trend break dated October 7, 2025, with a 95 percent confidence interval of 1.5x to 2.8x. The study ran 11 weeks from August 12 to October 29, 2026. METR estimates AI accounts for about 80 percent of the acceleration, with substantial uncertainty, and ruled out faster training-compute growth as the explanation. Anthropic’s Claude Opus 5.5 system card reported a 1.53x slope increase; the Haiku 5.5 card reported 1.7x. In a separate review the same day, METR called Anthropic’s claim it had not crossed its 2x Acceleration Threshold overstated and said crossing was plausible by July 15, while Anthropic’s August 2026 Risk Report covering through July 15 said acceleration was less than 2x. METR calculated about 1.97x from figures in Anthropic’s private report. Anthropic redacted most of METR’s key evidence in the public version of the roughly 100-page report, saying the cuts were aimed at removing confidential information that could accelerate other AI developers. On November 5, 2026, Sam McCandlish said Anthropic will publish its own pace-and-RSP-compliance analysis in coming weeks. Anthropic amended its IPO filing to say future models may soon meet the acceleration threshold.