| Benchmark | Claude Opus 4.8 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 49.4 | 14.4 |
| AA Intelligence | 57.3 | 7.7 |
| AA-LCR | 73 | 23 |
| AA-Omniscience | 28.8 | -68.4 |
| Artificial Analysis Coding Index | 74.3 | 6.7 |
| critpt | 20.9 | 0 |
| gdpval | 54.2 | 13.8 |
| GPQA Diamond | 93.6 | 64.1 |
| HLE | 57.9 | 4.6 |
| IFBench | 62.2 | 36.6 |
| include | 87.6 | 64.6 |
| mathvision | 86.7 | 60 |
| MMMU-Pro | 78.9 | 57 |
| OmniScience Accuracy | 48.8 | 12.2 |
| OmniScience Non-Hallucination | 60.7 | 8.2 |
| OSWorld-Verified | 83.4 | 31.4 |
| scicode | 53.5 | 17.1 |
| screenspot_pro_no_tools | 87.9 | 49.2 |
| Terminal-Bench Hard | 58.3 | 1.5 |
| Video-MME | 86 | 59.7 |
| τ²-Bench Telecom (AA run) | 94.4 | 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.