| Benchmark | DeepSeek-V4-Pro | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 53 | 10 |
| AA-LCR | 70 | 20.3 |
| AA-Omniscience | -10.6 | -52.2 |
| Artificial Analysis Coding Index | 59.4 | 11.9 |
| Chinese SimpleQA (C-SimpleQA) | 77.7 | 52.2 |
| Codeforces | 3206 | 24.8 |
| critpt | 13 | 0 |
| GPQA Diamond | 90.5 | 49.1 |
| HLE | 48.2 | 4.2 |
| IFBench | 76.5 | 36.9 |
| livebench | 73.6 | 52.3 |
| LiveCodeBench | 93.5 | 55.5 |
| MMLU-Pro | 87.5 | 71.6 |
| OmniScience Accuracy | 43 | 17.6 |
| OmniScience Non-Hallucination | 12.2 | 15.3 |
| scicode | 50 | 26.7 |
| simpleqa | 57.9 | 10.3 |
| SWE-bench Verified | 80.6 | 23.8 |
| Terminal-Bench Hard | 46.2 | 4.5 |
| τ²-Bench Telecom (AA run) | 96.2 | 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.