| Benchmark | o1 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 31.1 | 14.4 |
| AA Intelligence | 24 | 7.7 |
| AA-LCR | 63.3 | 23 |
| AA-Omniscience | -10.6 | -68.4 |
| AIME 2024 | 79.2 | 72.9 |
| AIME 2025 | 81.7 | 74.5 |
| Artificial Analysis Coding Index | 39.7 | 6.7 |
| critpt | 0.3 | 0 |
| gdpval | 11.5 | 13.8 |
| GPQA Diamond | 75.7 | 64.1 |
| HLE | 7.7 | 4.6 |
| IFBench | 70.3 | 36.6 |
| MathVista | 71.8 | 79.5 |
| OmniScience Accuracy | 34.7 | 12.2 |
| OmniScience Non-Hallucination | 30.7 | 8.2 |
| scicode | 35.8 | 17.1 |
| Terminal-Bench Hard | 12.9 | 1.5 |
| τ²-Bench Telecom (AA run) | 62.6 | 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.