METR Q4 2026 Time Horizon - TH2 Verifiable (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. Fitted only on the automatically scored half of TH2, METR's new suite: optimization contests, ML training, puzzles and games, plus the under-10-minute TH1.1 tasks METR adds to pin the short end. Much lower than the pooled series for the same model. Unbounded; higher is better. Reasoning effort and scaffold are not stated per model.

Metric: 50% Time Horizon (hours). Source: metr.org. Saturation forecast: Around December 2027. 11 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-6 Astra27.1#10
2Claude Opus 5.519.9#9
3Claude Fable 5.114.95#20
4Claude Opus 510.12#26
5GPT-5.56.17#54
6Claude Fable 56.13#33
7Claude Opus 4.83.61#65
8Claude Opus 4.72.51#85
9GPT-5.6 Sol2.35#37
10GPT-5.41.39#135
11Claude Opus 4.61.07#113

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-th2-verifiable-50 · How It Works · Data refreshed daily, snapshot 2026-11-21.