METR reports Anthropic model improvement has roughly doubled in a year

single source· 1 articles · confidence: low · first seen 2026-11-07 12:30 UTC

What this means for you

Nothing to act on yet. The claim is about how fast Anthropic's own models improve, not about anything you can call or deploy. Wait for METR's measurements — which models, which tasks, which metric — before citing the doubling.

Anthropic's models are improving roughly twice as quickly as they were a year ago, with little sign of the pace slowing, according to a report from METR, an AI safety research organisation, following an 11-week investigation. The report examines Anthropic's use of AI to do the work of building more capable AI. This account does not give the models tested, the tasks used or the metric behind the doubling, so the figure cannot yet be checked against a published evaluation.

Key facts

  • ·METR, an AI safety research organisation, published the report after an 11-week investigation. source
  • ·Anthropic's models were improving roughly twice as quickly as a year earlier. source
  • ·The report found little indication that the rate of improvement is slowing. source
  • ·The report concerns Anthropic's use of AI to build more powerful AI. source

What the sources say

  • The Verge AI — Single report on METR's 11-week probe into how much of Anthropic's development work its own models now do.

Sources

The original reporting. Follow these — they did the work.

← the wire