| Benchmark | DeepSeek-V3.2 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 32.8 | 10 |
| AA-LCR | 70.7 | 20.3 |
| AA-Omniscience | -22.5 | -52.2 |
| Artificial Analysis Coding Index | 44.2 | 11.9 |
| critpt | 2.9 | 0 |
| GPQA Diamond | 84 | 49.1 |
| HLE | 40.8 | 4.2 |
| IFBench | 61 | 36.9 |
| LiveCodeBench | 86 | 55.5 |
| longbench_v2 | 59.8 | 39.4 |
| mmlu_redux | 93.7 | 86.8 |
| MMLU-Pro | 86 | 71.6 |
| OmniScience Accuracy | 33 | 17.6 |
| OmniScience Non-Hallucination | 17.3 | 15.3 |
| scicode | 39 | 26.7 |
| SWE-bench Verified | 73.1 | 23.8 |
| Terminal-Bench Hard | 35.6 | 4.5 |
| τ²-Bench Telecom (AA run) | 90.6 | 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.