METR's evaluation of AI agent capabilities across different time horizons — from short tasks (minutes) to long autonomous operations (hours). Tests sustained agentic performance.
This definition has not completed editorial and provenance review. Verify its metric, version, source, and evaluation configuration upstream before using it for a model-selection decision.
Uniquely measures how model performance degrades with task duration. Critical for evaluating autonomous agent viability.
Score is 0–1 ratio. Evaluation methodology evolves as METR updates criteria.
Higher is better. Performance ratio (0–1) across time horizons.
Score is 0–1 ratio.
| # | Model | Score |
|---|---|---|
| 1 | 99.9 | |
| 2 | 99.2 | |
| 3 | 98.1 | |
| 4 | 81.1 | |
| 5 | 75.1 | |
| 6 | 73.6 | |
| 7 | 70.1 | |
| 8 | 69.9 | |
| 9 | 69.9 | |
| 10 | 60.5 | |
| 11 | 58.8 | |
| 12 | 57.3 | |
| 13 | 56.1 | |
| 14 | 55.3 | |
| 15 | 54.1 | |
| 16 | 53.9 | |
| 17 | 51.9 | |
| 18 | 49.7 | |
| 19 | 49.4 | |
| 20 | 49.2 | |
| 21 | 46.2 | |
| 22 | 46.1 | |
| 23 | 46.0 | |
| 24 | 45.1 | |
| 25 | 44.2 | |
| 26 | 43.1 | |
| 27 | 42.6 | |
| 28 | 42.5 | |
| 29 | 42.5 | |
| 30 | 42.0 | |
| 31 | 41.4 | |
| 32 | 41.3 | |
| 33 | 40.8 | |
| 34 | 40.3 | |
| 35 | 40.1 | |
| 36 | 40.0 | |
| 37 | 39.5 | |
| 38 | 38.3 | |
| 39 | 38.3 | |
| 40 | 38.3 | |
| 41 | 37.5 | |
| 42 | 37.2 | |
| 43 | 36.8 | |
| 44 | 36.0 | |
| 45 | 34.9 | |
| 46 | 34.3 | |
| 47 | 34.2 | |
| 48 | 32.4 | |
| 49 | 31.6 | |
| 50 | 31.1 |