| Benchmark | Claude Opus 4.5 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 59.6 | 14.4 |
| AA Intelligence | 41.9 | 7.7 |
| AA-LCR | 76 | 23 |
| AA-Omniscience | 14 | -68.4 |
| AIME 2025 | 92.8 | 74.5 |
| Artificial Analysis Coding Index | 47.8 | 6.7 |
| critpt | 4.6 | 0 |
| ERQA | 46.8 | 47.3 |
| gdpval | 47.3 | 13.8 |
| GPQA Diamond | 87 | 64.1 |
| HLE | 30.8 | 4.6 |
| IFBench | 58 | 36.6 |
| InfoVQA (val) | 76.9 | 79.5 |
| LiveCodeBench | 87 | 51.3 |
| LiveCodeBench v6 | 84.8 | 51.3 |
| LongVideoBench | 67.2 | 57.7 |
| mathvision | 77.1 | 60 |
| MathVista | 80.2 | 79.5 |
| mmlu_redux | 95.6 | 86 |
| MMLU-Pro | 90 | 73.6 |
| MMMU-Pro | 74 | 57 |
| MMStar | 73.2 | 73.2 |
| MV-Bench | 67.2 | 69.3 |
| OCRBench | 86.5 | 81.6 |
| OmniScience Accuracy | 46.6 | 12.2 |
| OmniScience Non-Hallucination | 39 | 8.2 |
| OSWorld-Verified | 66.3 | 31.4 |
| RealWorldQA | 77 | 73.2 |
| scicode | 50 | 17.1 |
| supergpqa | 70.6 | 46.8 |
| Terminal-Bench Hard | 47 | 1.5 |
| VideoMMMU | 84.4 | 69.4 |
| τ²-Bench Telecom (AA run) | 89.5 | 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.