Data behind: METR 50% time horizon of frontier AI agents (state-of-the-art models)

← Back · Data behind a figure on this site

METR 50% time horizon of frontier AI agents (state-of-the-art models)

How it's computed: Length of software/research tasks (measured in how long they take skilled humans) that a model completes with 50% (or 80%) reliability. Point estimates only; METR publishes wide confidence intervals. Each point is a model METR flags as state-of-the-art at release. Values exactly as published by the source (rounded).
Source: METR, Time Horizon v1.1 benchmark results · link [2] [3]
As of: 7 Apr 2026 · fetched 24 Oct 2026

18 rows

Year / date	50% time horizon (min) (minutes of human-expert task time)	80% time horizon (min) (minutes of human-expert task time)
2019-02-14	0.1	0
2020-05-28	0.1	0.1
2022-03-15	0.6	0.3
2023-03-14	4	0.9
2023-11-06	4	0.8
2024-05-13	7	1.3
2024-06-20	11.4	1.7
2024-09-12	20.3	4.4
2024-10-22	20.5	2.6
2024-12-05	38.8	7.1
2025-02-24	60.4	12.1
2025-04-16	119.7	30
2025-08-07	203	38.3
2025-11-18	224.3	54.1
2025-11-24	293	49.4
2025-12-11	352.2	66
2026-02-05	718.8	69.9
2026-04-07	1,044.8	185.9