Timeline Evaluation SIU-LDG-2025-003 SI Index +1.25

METR finds AI task horizons doubling about every seven months

The length of task frontier agents complete half the time has doubled roughly every seven months for six years. Claude 3.7 Sonnet is measured at about an hour.

Log-scale chart of METR time horizons rising from seconds in 2019 to about one hour for Claude 3.7 Sonnet in early 2025
METR 50% time horizons by release date, through Claude 3.7 Sonnet. Points show METR's later re-measurement on its Time Horizon 1.1 suite. Chart: SI USA, from METR data

Models succeed almost always on tasks that take humans under four minutes and rarely on tasks over four hours. The 50% crossover point is the “time horizon”. Extrapolated, the doubling trend implies week-long autonomous work within a few years. The Index does not extrapolate: each horizon milestone moves only when a measured point estimate crosses it.

Index treatment. Autonomous task horizon of one hour moves to 0.75.