| Benchmark | Gemini 3 Pro | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 40.6 | 10 |
| AA-LCR | 73 | 20.3 |
| AA-Omniscience | 15.3 | -52.2 |
| Artificial Analysis Coding Index | 46.5 | 11.9 |
| critpt | 9.1 | 0 |
| Fortress | 41.7 | 56.4 |
| GPQA Diamond | 91.9 | 49.1 |
| HLE | 45.8 | 4.2 |
| IFBench | 70.4 | 36.9 |
| livebench | 73.4 | 52.3 |
| LiveCodeBench | 92 | 55.5 |
| longbench_v2 | 68.2 | 39.4 |
| MMLU-Pro | 90.1 | 71.6 |
| mmmlu | 91.8 | 74.8 |
| OmniScience Accuracy | 55.8 | 17.6 |
| OmniScience Non-Hallucination | 10 | 15.3 |
| scicode | 56.1 | 26.7 |
| simpleqa | 72.1 | 10.3 |
| SWE-bench Verified | 78 | 23.8 |
| Terminal-Bench Hard | 56.9 | 4.5 |
| τ²-Bench Telecom (AA run) | 98 | 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.