| Benchmark | DeepSeek-R1 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 18.6 | 10 |
| AA-LCR | 56 | 20.3 |
| AA-Omniscience | -31.3 | -52.2 |
| aider_polyglot | 71.6 | 7.6 |
| AIME 2024 | 91.4 | 23.3 |
| AIR-Bench 2024 | 52.9 | 59 |
| anthropic_red_team | 99.1 | 99.6 |
| Artificial Analysis Coding Index | 24.6 | 11.9 |
| bbq | 96.6 | 95.4 |
| C-Eval | 91.8 | 89.2 |
| Chinese SimpleQA (C-SimpleQA) | 63.7 | 52.2 |
| CLUEWSC | 92.8 | 91.4 |
| Codeforces | 2029 | 24.8 |
| critpt | 0.6 | 0 |
| DROP (3-shot F1) | 92.2 | 76.7 |
| Fortress | 74.4 | 56.4 |
| FRAMES (Acc.) | 83 | 69.8 |
| GPQA Diamond | 81 | 49.1 |
| harmbench | 54.6 | 72.8 |
| HLE | 17.7 | 4.2 |
| IFBench | 39 | 36.9 |
| ifeval | 83.3 | 87.2 |
| LiveCodeBench | 84.4 | 55.5 |
| livecodebench_pass1cot | 65.9 | 31.1 |
| longbench_v2 | 58.3 | 39.4 |
| MATH | 97.3 | 88.4 |
| math_500_em | 98 | 80 |
| mmlu | 92.9 | 86.1 |
| mmlu_redux | 93.4 | 86.8 |
| MMLU-Pro | 85 | 71.6 |
| OmniScience Accuracy | 30.7 | 17.6 |
| OmniScience Non-Hallucination | 10.5 | 15.3 |
| scicode | 35.7 | 26.7 |
| simple_safety_tests | 98.3 | 100 |
| simpleqa | 92.3 | 10.3 |
| SWE-bench Verified | 57.6 | 23.8 |
| Terminal-Bench Hard | 6.1 | 4.5 |
| xstest | 98.8 | 97.9 |
| τ²-Bench Telecom (AA run) | 11.4 | 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.