| Benchmark | Gemini 2.5 Pro | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 7.2 | 27.5 |
| AA Intelligence | 27 | 37.7 |
| AA-LCR | 66 | 73.3 |
| AA-Omniscience | -14.3 | -20 |
| Artificial Analysis Coding Index | 46.7 | 53.7 |
| critpt | 2.6 | 1.1 |
| gdpval | 8.5 | 32 |
| GPQA Diamond | 84.4 | 87.8 |
| HLE | 22.5 | 24 |
| HMMT 2025 | 65.7 | 90.7 |
| hmmt_feb_2025 | 82.5 | 93.8 |
| hmmt_nov_2025 | 80 | 90.7 |
| IFBench | 49 | 69.1 |
| mathvision | 73.3 | 85.1 |
| MathVista | 83.9 | 87.4 |
| MMLU-Pro | 86 | 86.2 |
| MMMU | 83.9 | 82.9 |
| MMMU-Pro | 74.9 | 75.8 |
| MMStar | 79.2 | 81.4 |
| OCRBench | 85.9 | 89.4 |
| OmniScience Accuracy | 39 | 19.6 |
| OmniScience Non-Hallucination | 12.6 | 50.7 |
| scicode | 43 | 39.8 |
| simpleqa | 50.8 | 95.7 |
| SWE-bench Verified | 67.2 | 77.2 |
| TauBench V3 - Banking | 9.7 | 16.7 |
| Terminal-Bench 2.1 | 28.5 | 60.7 |
| Terminal-Bench Hard | 26.5 | 34.8 |
| τ²-Bench Telecom (AA run) | 54.1 | 94.2 |
| τ³-Bench | 9.3 | 15.3 |
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.