The paper recomputes time horizons on 228 tasks and 26 AIs and finds marked nonlinearity in the two-to-30-minute region.
Study finds equal-looking AI time-horizon gains can hide unequal difficulty
2026-10-09Frontier watchWATCH
A statistical reanalysis recomputes METR 50% time horizons across 228 tasks and 26 AIs. Nonlinear models show that moving from three to 30 minutes is much easier than moving from 30 minutes to five hours, despite both being tenfold gains. The authors recommend interpreting horizons together with diagnostic plots.
Availability: paper only; no public code or data package found
What this means for you
For teams using time horizons to plan agent roadmaps, a single exponential curve may overstate progress in the short-task region. This is a preprint without a public replication package; wait for data and code reproduction before changing procurement or automation assumptions.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Sample | 228 tasks and 26 AI systems |
|---|---|
| Methods | splines and item-response theory |
| Finding | human-time-to-AI-difficulty curve is nearly flat from 2 to 30 minutes |
| Availability | paper only; no public code or data package found |
Technical impact 3
Engineering value 3
Adoption signal 1
Maturity 1
Verification cost 4
Risk 2