| Benchmark | Gemma 4 31B | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 29.7 | 10 |
| AA-LCR | 68.3 | 20.3 |
| AA-Omniscience | -47.9 | -52.2 |
| Artificial Analysis Coding Index | 43.4 | 11.9 |
| C-Eval | 82.6 | 89.2 |
| critpt | 1.4 | 0 |
| GPQA Diamond | 85.7 | 49.1 |
| HLE | 26.5 | 4.2 |
| IFBench | 75.6 | 36.9 |
| mmlu_redux | 93.7 | 86.8 |
| MMLU-Pro | 85.2 | 71.6 |
| mmmlu | 88.4 | 74.8 |
| OmniScience Accuracy | 20 | 17.6 |
| OmniScience Non-Hallucination | 18.1 | 15.3 |
| scicode | 43.4 | 26.7 |
| SWE-bench Verified | 52 | 23.8 |
| Terminal-Bench Hard | 36.4 | 4.5 |
| τ²-Bench Telecom (AA run) | 65.5 | 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.