| Benchmark | Gemini 2.5 Pro | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 7.2 | 14.4 |
| AA Intelligence | 27 | 7.7 |
| AA-LCR | 66 | 23 |
| AA-Omniscience | -14.3 | -68.4 |
| AIME 2025 | 88.3 | 74.5 |
| Artificial Analysis Coding Index | 46.7 | 6.7 |
| critpt | 2.6 | 0 |
| gdpval | 8.5 | 13.8 |
| GPQA Diamond | 84.4 | 64.1 |
| HLE | 22.5 | 4.6 |
| HMMT 2025 | 65.7 | 53.1 |
| IFBench | 49 | 36.6 |
| LiveCodeBench | 82.7 | 51.3 |
| mathvision | 73.3 | 60 |
| MathVista | 83.9 | 79.5 |
| MMLU-Pro | 86 | 73.6 |
| MMMU-Pro | 74.9 | 57 |
| MMStar | 79.2 | 73.2 |
| OCRBench | 85.9 | 81.6 |
| OmniScience Accuracy | 39 | 12.2 |
| OmniScience Non-Hallucination | 12.6 | 8.2 |
| scicode | 43 | 17.1 |
| Terminal-Bench Hard | 26.5 | 1.5 |
| Video-MME | 84.8 | 59.7 |
| τ²-Bench Telecom (AA run) | 54.1 | 15.5 |
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.