METR estimates Anthropic's model progress rate has doubled

METR said in an assessment that Anthropic is likely improving model capabilities on well-specified tasks about twice as quickly as it did from March 2024 to October 2025, though the true figure could be closer to 1.3 or 3 times the historical rate. The evaluator reviewed internal research data and interviewed staff from August 12 to October 29, 2026.

Anthropic's internal AECI benchmark scores grew at an estimated 1.9 times the earlier rate after October 2025, with a 95% confidence interval of 1.5 to 2.8 times. METR said this amounted to roughly two years of historical progress within one year and concluded that faster growth in training compute did not cause the acceleration.

METR estimated that AI accounts for a majority of the current acceleration, possibly about 80%, while noting substantial uncertainty. Staff reported that AI mainly assists researchers rather than completing work autonomously. METR saw no evidence of significant bottlenecks in algorithms, compute or data over the next 12 months but said it is not confident in that assessment, and Anthropic redacted some findings that informed METR's conclusions.