METR Q4 2026 Time Horizon - MirrorCode (80%): leaderboard
METR's Q4 2026 capability update (20 November 2026): the length of task, in hours of expert human work, that a model completes 80% of the time, fitted across tasks of different lengths. Fitted only on MirrorCode's 75 software replication tasks (25 command-line programs, each rebuilt in three languages), from older runs that stop at GPT-5.4 and Claude Opus 4.6. Unbounded; higher is better. Reasoning effort and scaffold are not stated per model.
Metric: 80% Time Horizon (hours). Source: metr.org. Saturation forecast: Around June 2027. 6 models tracked. Updated 21 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 77.68 | #113 |
| 2 | GPT-5.4 | 29.7 | #135 |
| 3 | GPT-5.2 | 12.89 | #185 |
| 4 | Claude Opus 4.5 | 6.71 | #160 |
| 5 | GPT-5 | 4.24 | #178 |
| 6 | Claude Opus 4.1 | 4 | #278 |
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-mirrorcode-80 · How It Works · Data refreshed daily, snapshot 2026-11-21.