METR said its August–October 2026 high-access assessment found Anthropic’s AECI benchmark scores growing at about 1.9x the pre-October 2025 rate, with confidence the acceleration was not from faster training-compute scaling. METR judged a majority of the current acceleration likely comes from using AI in R&D. The Verge reported the same roughly 2x finding and noted Anthropic redacted much of the most informative evidence from the public report.