METR Estimates Anthropic's AI Model Capability Gains Have Nearly Doubled in Speed

At a glance

The independent evaluation group METR believes Anthropic's AI models are likely improving their abilities on clearly defined tasks about twice as fast as before. It judged that much of this speedup appears to come from using AI in research. However, the conclusion has large uncertainty, and much of the supporting data was not made public.

Key points

1. 1

   METR estimated that benchmark scores have risen about 1.9 times faster than before since October 2025.

2. 2

   The effect of AI helping researchers appeared to be the main cause of the current speedup.

3. 3

   Employee responses were mixed, and self-reported productivity gains may not represent the whole picture.

4. 4

   Anthropic removed data it did not make public, so a majority of the evidence was left out of the report.

Why it matters

As AI development speeds up, its effects on society can also grow quickly, which makes this important.