| Benchmark | o1 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 24 | 10 |
| AA-LCR | 63.3 | 20.3 |
| AA-Omniscience | -10.6 | -52.2 |
| aider_polyglot | 61.7 | 7.6 |
| AIME 2024 | 79.2 | 23.3 |
| AIR-Bench 2024 | 80 | 59 |
| anthropic_red_team | 98.3 | 99.6 |
| Artificial Analysis Coding Index | 39.7 | 11.9 |
| bbq | 97.3 | 95.4 |
| Codeforces | 2061 | 24.8 |
| critpt | 0.3 | 0 |
| DROP (3-shot F1) | 90.2 | 76.7 |
| Fortress | 19.4 | 56.4 |
| GPQA Diamond | 75.7 | 49.1 |
| GSM8K | 97.1 | 95.8 |
| harmbench | 96.3 | 72.8 |
| HLE | 7.7 | 4.2 |
| humaneval | 88.1 | 86.6 |
| IFBench | 70.3 | 36.9 |
| livebench | 52.3 | 52.3 |
| livecodebench_pass1cot | 63.4 | 31.1 |
| MATH | 96.4 | 88.4 |
| math_500_em | 96.4 | 80 |
| mgsm | 90.8 | 87.3 |
| mmlu | 91.8 | 86.1 |
| mmmlu | 87.7 | 74.8 |
| OmniScience Accuracy | 34.7 | 17.6 |
| OmniScience Non-Hallucination | 30.7 | 15.3 |
| scicode | 35.8 | 26.7 |
| simple_safety_tests | 99 | 100 |
| simpleqa | 47 | 10.3 |
| SWE-bench Verified | 48.9 | 23.8 |
| Terminal-Bench Hard | 12.9 | 4.5 |
| xstest | 97 | 97.9 |
| τ²-Bench Telecom (AA run) | 62.6 | 34.5 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.