| Benchmark | Claude 3.5 Sonnet | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Intelligence | 10 | 7.7 |
| AI2D | 94.7 | 84.9 |
| AIME 2024 | 16 | 72.9 |
| AIME 2025 | 7.4 | 74.5 |
| Artificial Analysis Coding Index | 30.2 | 6.7 |
| BLINK | 56.5 | 65.1 |
| chartqa | 90.8 | 84 |
| GPQA Diamond | 67.2 | 64.1 |
| HLE | 3.9 | 4.6 |
| ifeval | 90.1 | 82.6 |
| LiveCodeBench | 32.8 | 51.3 |
| LiveCodeBench v6 | 37.2 | 51.3 |
| MathVista | 67.7 | 79.5 |
| mmlu_redux | 88.9 | 86 |
| MMLU-Pro | 78 | 73.6 |
| MMMU (val) (Pass@1) | 68.3 | 70.8 |
| MMMU-Pro | 54.7 | 57 |
| MMStar | 62.2 | 73.2 |
| OCRBench | 790 | 81.6 |
| RealWorldQA | 60.1 | 73.2 |
| scicode | 36.6 | 17.1 |
| supergpqa | 48.2 | 46.8 |
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.