# METR Time Horizons

Measures how long a task (in human working time) an AI can complete about half the time; shows how fast agents are improving.

Category: agents. Type: leaderboard. Status: active.
Metric: 50% time horizon, minutes (95% CI), as in METR data file (estimate field; unit assumed minutes) (higher is better).
Source: METR — https://metr.org/assets/benchmark_results_1_1.yaml, results as of 2026-09-08, retrieved 2026-11-17, license: not stated.

| Rank | Model | 50% time horizon (min, CI) | Release date |
|---|---|---|---|
| 1 | claude_mythos_preview_early_inspect | 1044.8 min (508.9-3304.3) | 2026-04-07 |
| 2 | claude_opus_4_6_inspect | 718.8 min (316.7-3633.8) | 2026-02-05 |
| 3 | gemini_3_1_pro | 384.1 min (233.5-694.8) | 2026-02-19 |
| 4 | gpt_5_2 | 352.2 min (198.1-815.2) | 2025-12-11 |
| 5 | gpt_5_3_codex | 349.5 min (194.9-816.4) | 2026-02-05 |
| 6 | gpt_5_4 | 341.7 min (186.6-768.8) | 2026-03-05 |
| 7 | claude_opus_4_5_inspect | 293.0 min (161.7-623.7) | 2025-11-24 |
| 8 | gemini_3_pro | 224.3 min (139.6-379.2) | 2025-11-18 |
| 9 | gpt_5_1_codex_max_inspect | 223.7 min (134.3-396.2) | 2025-11-19 |
| 10 | gpt_5_2025_08_07_inspect | 203.0 min (112.6-405.6) | 2025-08-07 |
| 11 | o3_inspect | 119.7 min (74.6-190.9) | 2025-04-16 |
| 12 | claude_4_1_opus_inspect | 100.5 min (59.3-159.5) | 2025-08-05 |
| 13 | claude_4_opus_inspect | 100.4 min (60.0-163.5) | 2025-05-22 |
| 14 | claude_3_7_sonnet_inspect | 60.4 min (33.0-104.2) | 2025-02-24 |
| 15 | o1_inspect | 38.8 min (21.2-65.0) | 2024-12-05 |
| 16 | claude_3_5_sonnet_20241022_inspect | 20.5 min (10.1-40.8) | 2024-10-22 |
| 17 | o1_preview | 20.3 min (11.7-33.4) | 2024-09-12 |
| 18 | claude_3_5_sonnet_20240620_inspect | 11.4 min (5.5-22.4) | 2024-06-20 |
| 19 | gpt_4o_inspect | 7.0 min (4.0-12.9) | 2024-05-13 |
| 20 | gpt_4_1106_inspect | 4.0 min (1.9-8.4) | 2023-11-06 |
| 21 | gpt_4 | 4.0 min (1.9-8.0) | 2023-03-14 |
| 22 | claude_3_opus_inspect | 4.0 min (1.7-8.8) | 2024-03-04 |
| 23 | gpt_4_turbo_inspect | 3.7 min (2.0-6.7) | 2024-04-09 |
| 24 | gpt_3_5_turbo_instruct | 0.6 min (0.3-1.1) | 2022-03-15 |
| 25 | davinci_002 | 0.1 min (0.1-0.2) | 2020-05-28 |

Notes: METR-Horizon-v1.1 data file, top 25 sorted by p50 estimate (rank is mine, not source order). Page says it is no longer actively updated; unit minutes inferred (GPT-5 ~203 min vs. page's ~2h17m).
