Benchmarks / Agentic
METR Time Horizon (50%)
METR's time horizon: the length of software and reasoning task, measured in minutes a skilled human needs, that the model completes successfully half the time. Longer is better; it has roughly doubled every few months.
METR Time Horizon (50%)
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Mythos Preview | – | 1045 |
| 2 | Opus 4.6 | max | 719 |
| 3 | Gemini 3.1 Pro | high | 384 |
| 4 | GPT-5.2 | xhigh | 352 |
| 5 | GPT-5.3 Codex | xhigh | 350 |
| 6 | GPT-5.4 | xhigh | 342 |
| 7 | Opus 4.5 | high | 293 |
| 8 | Gemini 3 Pro | high | 224 |
| 9 | GPT-5.1 Codex | max | 224 |
| 10 | GPT-5.2025 | high | 203 |
| 11 | o3 | high | 120 |
| 12 | Opus 4.1 | high | 100 |
| 13 | Opus 4 | medium | 100 |
| 14 | Sonnet 3.7 | medium | 60 |
| 15 | o1 | high | 39 |
| 16 | GPT-4o | high | 7 |
| 17 | GPT-4.1106 | – | 4 |
| 18 | GPT-4 | high | 4 |
| 19 | Opus 3 | none | 4 |
| 20 | GPT-3.5 | high | 1 |
| 21 | GPT-2 | – | 0 |