METR: Anthropic's AI R&D pace likely about 2x, says public report "understates" acceleration
First seen on X 6 hours ago@METR_Evals ♥ 2,0501 views
Is AI speeding up AI development? METR answered on Nov. 5 after working since August with internal Anthropic information. It submitted a roughly 100-page report to Anthropic; for the public version, Anthropic redacted most of the evidence METR found most informative, and METR said the public must rely on its judgment more than ideal.
METR judged that as of October 2026 the pace of capability growth on well-specified tasks is likely about 2x what it was from March 2024 to October 2025, with 1.3x to 3x not surprising. A companion review covered the 'Risks from automated R&D' section of Anthropic's August risk report. METR agreed catastrophic risk from July 15 models was low, but called the claim that the RSP's '2x acceleration' threshold was not crossed overconfident and said it may have been crossed. Most METR and Anthropic researchers lean toward 'not crossed', with low confidence.
The key dispute is wording. The public report said progress was 'less than' 2x; METR computed 1.97 from private figures and said 'around 2' fits better, with its own range at 1.3 to 2.8. Anthropic itself said it is less confident than before because its most concrete evaluations have saturated and early signs of acceleration are visible.
Andrew Curran summarized it as METR saying Anthropic has probably hit the threshold (about 550 likes), but METR wrote that it is plausible, not confirmed. With Anthropic heading to an IPO, attention is on how investors take it.
“Since August, a team within METR has worked with internal Anthropic information to assess whether AI itself is accelerating AI development.”
