THE EVIDENCE LEDGER
How did METR’s task-horizon method change?
A 50% task-success horizon is not reliable unsupervised work. Confidence intervals are wide.
What was claimed?
METR’s TH1.1 revision improves its task suite and changes some model estimates.
What does the evidence support?
Independent evidence, scope limited
Evidence supports progress on software and research tasks. Measurement changes also affect the apparent growth rate.
What are the limits?
A 50% task-success horizon is not reliable unsupervised work. Confidence intervals are wide.
Claim or evidence date: 2026-01-29. Last verified: .
Colors classify evidence; they do not rank truthfulness. Explore hype versus reality →