| Benchmark | DeepSeek-V4-Flash | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 51.8 | 10 |
| AA-LCR | 74.3 | 20.3 |
| AA-Omniscience | -14.3 | -52.2 |
| Artificial Analysis Coding Index | 69.1 | 11.9 |
| Chinese SimpleQA (C-SimpleQA) | 78.9 | 52.2 |
| Codeforces | 3052 | 24.8 |
| critpt | 16.6 | 0 |
| GPQA Diamond | 90.8 | 49.1 |
| HLE | 38.6 | 4.2 |
| IFBench | 79.2 | 36.9 |
| LiveCodeBench | 91.6 | 55.5 |
| MMLU-Pro | 86.4 | 71.6 |
| OmniScience Accuracy | 40.4 | 17.6 |
| OmniScience Non-Hallucination | 15.6 | 15.3 |
| scicode | 49.9 | 26.7 |
| SWE-bench Verified | 79 | 23.8 |
| Terminal-Bench Hard | 38.6 | 4.5 |
| τ²-Bench Telecom (AA run) | 95.6 | 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.