| Benchmark | o3 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 36.1 | 14.4 |
| AA Intelligence | 31.1 | 7.7 |
| AA-LCR | 73.3 | 23 |
| AA-Omniscience | -15.3 | -68.4 |
| AIME 2024 | 91.6 | 72.9 |
| AIME 2025 | 89.2 | 74.5 |
| Artificial Analysis Coding Index | 38.4 | 6.7 |
| critpt | 1.1 | 0 |
| ERQA | 64 | 47.3 |
| gdpval | 12.8 | 13.8 |
| GPQA Diamond | 83.3 | 64.1 |
| HLE | 20.3 | 4.6 |
| IFBench | 71.4 | 36.6 |
| LiveCodeBench | 84.7 | 51.3 |
| MathVista | 86.8 | 79.5 |
| MMLU-Pro | 85 | 73.6 |
| MMMU-Pro | 76.4 | 57 |
| OmniScience Accuracy | 38.6 | 12.2 |
| OmniScience Non-Hallucination | 12.9 | 8.2 |
| scicode | 41 | 17.1 |
| Terminal-Bench Hard | 37.1 | 1.5 |
| VideoMMMU | 83.3 | 69.4 |
| τ²-Bench Telecom (AA run) | 80.7 | 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.