Anthropic's rate of model capability improvement has roughly doubled compared with the pace observed from March 2024 through October 2025, according to an independent assessment published by METR on November 5, 20262. The AI research nonprofit estimates that internal AI use within Anthropic explains a majority of the acceleration1.
Using Anthropic's benchmark capability index (AECI), METR estimated that benchmark scores have grown 1.9 times faster since October 2025, with a 95% confidence interval of 1.5 to 2.8 times. At that rate, roughly two years of progress at the previous pace would be compressed into a single year.
The calculation includes internal-only models and dates each snapshot by when most Anthropic employees could use it, creating a common basis for released and unreleased systems. Robustness checks produced acceleration estimates ranging from 1.1 to 2.9 times under alternative modeling choices, reflecting sensitivity to how benchmark difficulty and other analytical assumptions are represented.
METR's conclusions carry substantial uncertainty, and redactions conceal much of the supporting evidence. The assessment's broader judgment compares October 2026 performance with the rate observed from March 2024 through October 2025.
Internal AI use drove about 80% of the acceleration, according to the report. METR found that faster growth in training compute did not cause the speedup, though the supporting details were redacted. Non-AI factors likely played a large role in late 2025, but METR's current best guess is that they do not explain the sustained acceleration.
ANALYSIS The finding that a frontier lab's own AI tools are now the primary engine behind its capability gains marks a concrete data point for recursive improvement dynamics — the scenario in which AI systems accelerate the development of their successors. The wide confidence interval and redacted evidence, however, limit how firmly the magnitude can be pinned down.