| Benchmark | Gemini 3 Pro | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 52 | 14.4 |
| AA Intelligence | 40.6 | 7.7 |
| AA-LCR | 73 | 23 |
| AA-Omniscience | 15.3 | -68.4 |
| AIME 2025 | 100 | 74.5 |
| Artificial Analysis Coding Index | 46.5 | 6.7 |
| critpt | 9.1 | 0 |
| gdpval | 34.2 | 13.8 |
| GPQA Diamond | 91.9 | 64.1 |
| HLE | 45.8 | 4.6 |
| IFBench | 70.4 | 36.6 |
| InfoVQA (val) | 57.2 | 79.5 |
| LiveCodeBench | 92 | 51.3 |
| LiveCodeBench v6 | 90.7 | 51.3 |
| LongVideoBench | 77.7 | 57.7 |
| LVBench | 73.5 | 53.5 |
| mathvision | 86.1 | 60 |
| MathVista | 89.8 | 79.5 |
| MMLU-Pro | 90.1 | 73.6 |
| MMMU-Pro | 81 | 57 |
| OCRBench | 90.3 | 81.6 |
| OmniScience Accuracy | 55.8 | 12.2 |
| OmniScience Non-Hallucination | 10 | 8.2 |
| scicode | 56.1 | 17.1 |
| screenspot_pro_no_tools | 72.7 | 49.2 |
| Terminal-Bench Hard | 56.9 | 1.5 |
| Video-MME | 88.4 | 59.7 |
| VideoMMMU | 87.6 | 69.4 |
| τ²-Bench Telecom (AA run) | 98 | 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.