| Benchmark | Gemini 3.1 Pro | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 47.7 | 10 |
| AA-LCR | 79 | 20.3 |
| AA-Omniscience | 32.9 | -52.2 |
| Artificial Analysis Coding Index | 68.8 | 11.9 |
| Chinese SimpleQA (C-SimpleQA) | 85.9 | 52.2 |
| Codeforces | 3052 | 24.8 |
| critpt | 17.7 | 0 |
| Fortress | 29.8 | 56.4 |
| GPQA Diamond | 94.3 | 49.1 |
| HLE | 51.4 | 4.2 |
| IFBench | 77.1 | 36.9 |
| livebench | 79.9 | 52.3 |
| LiveCodeBench | 91.7 | 55.5 |
| mmlu | 92.6 | 86.1 |
| MMLU-Pro | 91 | 71.6 |
| mmmlu | 92.6 | 74.8 |
| OmniScience Accuracy | 55.3 | 17.6 |
| OmniScience Non-Hallucination | 50.1 | 15.3 |
| scicode | 59 | 26.7 |
| SWE-bench Verified | 80.6 | 23.8 |
| Terminal-Bench Hard | 68.5 | 4.5 |
| τ²-Bench Telecom (AA run) | 99.3 | 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.