METR Benchmark (80% Horizon): leaderboard
METR's 80% task-completion time horizon: the task length, in human-expert hours, that an AI agent can complete with roughly 80% reliability.
Metric: 80% Time Horizon (hours). Source: metr.org. Saturation forecast: Around August 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Mythos Preview | 3.1 | #4 |
| 2 | Gemini 3.1 Pro (Preview) | 1.5 | #58 |
| 3 | Claude Opus 4.6 | 1.16 | #63 |
| 4 | GPT-5.2 | 1.1 | #110 |
| 5 | GPT-5.3 Codex | 0.91 | #69 |
| 6 | GPT-5.4 | 0.9 | #79 |
| 7 | Gemini 3 Pro | 0.9 | #80 |
| 8 | GPT-5.1 Codex Max | 0.84 | #100 |
| 9 | Claude Opus 4.5 | 0.82 | #83 |
| 10 | GPT-5 | 0.64 | #98 |
| 11 | O3 | 0.5 | #124 |
| 12 | Claude Opus 4.1 | 0.39 | #140 |
| 13 | Claude Opus 4 | 0.34 | #163 |
| 14 | Claude 3.7 Sonnet | 0.2 | #249 |
| 15 | O1 | 0.12 | #183 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=metr-benchmark-80-horizon · How It Works · Data refreshed daily, snapshot 2026-10-20.