METR Says Anthropic Is Improving Twice as Fast
An outside evaluator puts AI’s own contribution at roughly 80% of the recent gain, with wide uncertainty attached.
The AI safety nonprofit METR published an assessment on 5 November 2026, drawing on about 11 weeks of work from August to October, finding that Anthropic’s model capabilities have been improving at roughly 1.9 times the earlier rate since October 2025. METR attributes about 80% of that acceleration to AI assistance in Anthropic’s own research, while saying the estimate carries substantial uncertainty. The finding rests on confidential research, interviews and unpublished model data, and independent replication would be difficult, which is precisely why it will be argued over.
