METR Q4 2026 Time Horizon - TH1.1 (50%): leaderboard

METR's Q4 2026 capability update (20 November 2026): the length of task, in hours of expert human work, that a model completes 50% of the time, fitted across tasks of different lengths. The post's own re-fit of the 222 Time Horizon 1.1 tasks alone. The numbers differ from METR's dashboard series (the METR Benchmark board), and at 50% the series stops at GPT-5.4 because later models pass the suite's 30-hour longest task. Unbounded; higher is better. Reasoning effort and scaffold are not stated per model.

Metric: 50% Time Horizon (hours). Source: metr.org. Saturation forecast: Around June 2027. 19 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-5.419.77#135
2GPT-5.3 Codex13.64#124
3GPT-5.210.53#185
4Gemini 3.1 Pro (Preview)8.05#99
5Claude Opus 4.58.02#160
6Gemini 3 Pro6.06#145
7GPT-55.29#178
8GPT-5.1 Codex Max5.27#207
9Claude Opus 4.12.87#278
10Claude Opus 42.65#326
11O32.18#237
12Claude 3.7 Sonnet1.46#470
13O10.972#403
14O1 Preview0.4058#467
15Claude 3.5 Sonnet (20241022)0.3495#648

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=metr-q4-2026-time-horizon-th1-1-50 · How It Works · Data refreshed daily, snapshot 2026-11-21.