| Benchmark | GPT-4o | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 9.7 | 14.4 |
| AA Intelligence | 11.2 | 7.7 |
| AA-LCR | 55 | 23 |
| AA-Omniscience | -10.5 | -68.4 |
| AI2D | 94.2 | 84.9 |
| AIME 2024 | 9.3 | 72.9 |
| AIME 2025 | 11.7 | 74.5 |
| Artificial Analysis Coding Index | 24.2 | 6.7 |
| chartqa | 88.1 | 84 |
| critpt | 0 | 0 |
| ERQA | 35.2 | 47.3 |
| gdpval | 0 | 13.8 |
| GPQA Diamond | 70.1 | 64.1 |
| HLE | 5.3 | 4.6 |
| IFBench | 36 | 36.6 |
| ifeval | 84.3 | 82.6 |
| LiveCodeBench | 38.3 | 51.3 |
| LiveCodeBench v6 | 30.9 | 51.3 |
| mathvision | 30.4 | 60 |
| MathVista | 63.8 | 79.5 |
| mmlu_prox | 61.1 | 65 |
| mmlu_redux | 88 | 86 |
| MMLU-Pro | 74.7 | 73.6 |
| MMMU (val) (Pass@1) | 69.1 | 70.8 |
| MMMU-Pro | 59.9 | 57 |
| MMStar | 63.9 | 73.2 |
| OCRBench | 806 | 81.6 |
| OmniScience Accuracy | 23.7 | 12.2 |
| OmniScience Non-Hallucination | 62.1 | 8.2 |
| OSWorld-Verified | 5 | 31.4 |
| RealWorldQA | 75.4 | 73.2 |
| scicode | 33.4 | 17.1 |
| supergpqa | 42.4 | 46.8 |
| Terminal-Bench Hard | 8.3 | 1.5 |
| Video-MME | 77.2 | 59.7 |
| VideoMMMU | 61.2 | 69.4 |
| VSI-Bench | 34 | 55.2 |
| τ²-Bench Telecom (AA run) | 28.9 | 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.