METR estimates Anthropic improves models on well-specified tasks twice as fast

TL;DR

  • From 12 August to 29 October 2026, METR assessed Anthropic from the inside and concluded that Anthropic is likely improving its models on well-specified tasks at around twice its March 2024 to October 2025 rate.
  • METR puts a majority of that on Anthropic's own use of AI in research, and it is confident that faster growth in training compute didn't cause it.
  • Anthropic redacted a majority of the evidence METR found most informative, and co-founder Sam McCandlish defended that on X.

Read the full story

Sign in with your email to read AI News. It’s free.

Share this story

Explain like I’m 15