See https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/

Resolves to the longest 80% Time Horizon, as measured by METR, for any AI system, by the end of September 2026. Answers that are passed early can be resolved early. The most recent suite of METR TH tasks will be used for resolution (whether v1.1, v2, v3, etc.).

IMPORTANT: Resolves to all thresholds reached, not just to the highest one that applies. eg for a 11 hour 80% time horizon, "≥10h" resolves yes, but so does ≥6h and ≥8h

Measuring AI Ability to Complete Long Tasks

We propose measuring AI performance in terms of the *length* of tasks AI agents can complete. We show that this metric has been consistently exponentially increasing over the past 6 years, with a doubling time of around 7 months. Extrapolating this trend predicts that, in under a decade, we will see AI agents that can independently complete a large fraction of software tasks that currently take humans days or weeks.

See also:

/Bayesian/best-metr-time-horizons-in-2026
/Bayesian/best-metr-80-time-horizon-before-au

/Bayesian/best-metr-80-time-horizon-before-oc