Benchmarks
METR Time Horizons
Models scored
50
Best score
0.85
Metric
average_score
METR Time Horizons is an AI evaluation tracked by Epoch AI, with 50 scored model versions on record. The scores here are reported by third parties rather than produced by Epoch AI running the evaluation itself.
Scored models span November 2019 to April 2026. The best score recorded is 0.85, measured as average_score. Models come from Anthropic, Google DeepMind, OpenAI, xAI, Moonshot, DeepSeek, Alibaba.
Full record
- Score metric
- average_score
- Models scored
- 50
- Best score recorded
- 0.85
- Earliest model scored
- Nov 5, 2019
- Most recent model scored
- Apr 7, 2026
- Third-party reported
- Yes
Leaderboard
| 01 | claude-mythos-preview-early | Anthropic | 0.85 |
| 02 | claude-opus-4-6_unknown | Anthropic | 0.79 |
| 03 | gemini-3.1-pro-preview | Google DeepMind | 0.77 |
| 04 | gpt-5.2-2025-12-11_high | OpenAI | 0.75 |
| 05 | claude-opus-4-5-20251101_16K | Anthropic | 0.75 |
| 06 | gpt-5.3-codex | OpenAI | 0.75 |
| 07 | gpt-5.4-2026-03-05_unknown | OpenAI | 0.74 |
| 08 | gpt-5.4-2026-03-05_xhigh | OpenAI | 0.74 |
| 09 | claude-opus-4-5-20251101 | Anthropic | 0.73 |
| 10 | gpt-5.1-codex-max | OpenAI | 0.72 |
| 11 | gemini-3-pro-preview | Google DeepMind | 0.71 |
| 12 | gpt-5-2025-08-07_medium | OpenAI | 0.7 |
| 13 | gpt-5-2025-08-07_high | OpenAI | 0.69 |
| 14 | claude-sonnet-4-5-20250929_16K | Anthropic | 0.67 |
| 15 | claude-opus-4-1-20250805_16K | Anthropic | 0.67 |
| 16 | grok-4-0709 | xAI | 0.67 |
| 17 | o3-2025-04-16_medium | OpenAI | 0.65 |
| 18 | claude-opus-4-20250514_16K | Anthropic | 0.64 |
| 19 | o4-mini-2025-04-16_medium | OpenAI | 0.64 |
| 20 | o3-2025-04-16_unknown | OpenAI | 0.64 |
| 21 | claude-sonnet-4-20250514_16K | Anthropic | 0.62 |
| 22 | claude-opus-4-1-20250805 | Anthropic | 0.62 |
| 23 | claude-opus-4-20250514_unknown | Anthropic | 0.62 |
| 24 | claude-3-7-sonnet-20250219_16K | Anthropic | 0.6 |
| 25 | kimi-k2-thinking | Moonshot | 0.59 |
| 26 | gpt-oss-120b | OpenAI | 0.57 |
| 27 | o1-2024-12-17_medium | OpenAI | 0.56 |
| 28 | claude-3-7-sonnet-20250219 | Anthropic | 0.56 |
| 29 | gemini-2.5-pro-preview-06-05 | Google DeepMind | 0.55 |
| 30 | DeepSeek-R1-0528 | DeepSeek | 0.54 |
| 31 | claude-3-5-sonnet-20241022 | Anthropic | 0.53 |
| 32 | DeepSeek-R1 | DeepSeek | 0.52 |
| 33 | DeepSeek-V3-0324 | DeepSeek | 0.5 |
| 34 | o1-preview-2024-09-12 | OpenAI | 0.49 |
| 35 | claude-3-5-sonnet-20240620 | Anthropic | 0.48 |
| 36 | DeepSeek-V3 | DeepSeek | 0.47 |
| 37 | gpt-4o-2024-11-20 | OpenAI | 0.41 |
| 38 | gpt-4-1106-preview | OpenAI | 0.4 |
| 39 | claude-3-opus-20240229 | Anthropic | 0.38 |
| 40 | gpt-4-turbo-2024-04-09 | OpenAI | 0.37 |
| 41 | gpt-4-0314 | OpenAI | 0.36 |
| 42 | qwen2.5-72b-instruct | Alibaba | 0.36 |
| 43 | gpt-4-0125-preview | OpenAI | 0.35 |
| 44 | gpt-4o-2024-08-06 | OpenAI | 0.34 |
| 45 | qwen2-72b-instruct | Alibaba | 0.3 |
| 46 | gpt-4-0613 | OpenAI | 0.29 |
| 47 | gpt-4-turbo | OpenAI | 0.27 |
| 48 | gpt-3.5-turbo-instruct | OpenAI | 0.21 |
| 49 | davinci-002 | OpenAI | 0.16 |
| 50 | gpt2-xl | OpenAI | 0.1 |
More benchmarks
SourceEpoch AI, 'AI Benchmarking Hub'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/benchmarks. Licensed under CC BY 4.0.
← All benchmarks