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.
21 models · 21 results · data as of 2026-10-29 · metr.org ↗
Leaderboard
Mythos Preview
All results
21 rows| # | Model | Effort | METR Time Horizon (50%) | Harness |
|---|---|---|---|---|
| 1 | Mythos Preview | 1044.8 ±1397.7 | METR | |
| 2 | Opus 4.6 | max | 718.8 ±1658.6 | METR |
| 3 | Gemini 3.1 Pro | high | 384.1 ±230.6 | METR |
| 4 | GPT-5.2 | xhigh | 352.2 ±308.6 | METR |
| 5 | GPT-5.3 Codex | xhigh | 349.5 ±310.7 | METR |
| 6 | GPT-5.4 | xhigh | 341.7 ±291.1 | METR |
| 7 | Opus 4.5 | high | 293.0 ±231.0 | METR |
| 8 | Gemini 3 Pro | high | 224.3 ±119.8 | METR |
| 9 | GPT-5.1 Codex | max | 223.7 ±130.9 | METR |
| 10 | GPT-5.2025 | high | 203.0 ±146.5 | METR |
| 11 | o3 | high | 119.7 ±58.2 | METR |
| 12 | Opus 4.1 | high | 100.5 ±50.1 | METR |
| 13 | Opus 4 | medium | 100.4 ±51.7 | METR |
| 14 | Sonnet 3.7 | medium | 60.4 ±35.6 | METR |
| 15 | o1 | high | 38.8 ±21.9 | METR |
| 16 | GPT-4o | high | 7.0 ±4.5 | METR |
| 17 | GPT-4.1106 | 4.0 ±3.3 | METR | |
| 18 | GPT-4 | high | 4.0 ±3.0 | METR |
| 19 | Opus 3 | none | 4.0 ±3.5 | METR |
| 20 | GPT-3.5 | high | 0.6 ±0.4 | METR |
| 21 | GPT-2 | 0.1 ±0.1 | METR |