Measuring AI Ability to Complete Long Tasks
The roughly seven-month doubling since 2019 with recent acceleration; about one hour for early-2025 frontier models; the climb toward twelve hours by early 2026.
Cited in
METR. 'Measuring AI Ability to Complete Long Tasks' (Kwa et al., 2025) and 'Time Horizon 1.1' (January 2026). (The roughly seven-month doubling since 2019 with recent acceleration; about one hour for early-2025 frontier models; the climb toward twelve hours by early 2026.)