Task reliability doubles every forty days

Claim

METR-style task reliability is doubling every 40 days, on track for one model to outwork every human who ever lived by 2032, while Benedict Evans argued tokens are destined to be low-margin commodity infrastructure.

Reported metric: doubling every 40 days

Reporting date: 2026-07-12 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

• Welcome to July 12, 2026 (theinnermostloop.substack.com)