METR offers a potentially useful evaluation frame, while its preliminary NanoGPT findings do not establish broad AI R&D acceleration.
Limits of the evidence: The article reports METR's own proposed measure and preliminary NanoGPT results; the source does not provide independent replication or evidence that the estimates generalize to other optimization tasks.
What to watch
Taken from the first report; the whole story has not been re-assessed yet.
Expenditure Horizon: Measuring Optimization Ability, with an Application to NanoGPT
METR BlogJul 21, 2026First report
No revisions yet.
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.