| Benchmark | o4-mini | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 36.1 | 14.4 |
| AA Intelligence | 26.1 | 7.7 |
| AA-LCR | 60 | 23 |
| AA-Omniscience | -35.7 | -68.4 |
| AIME 2025 | 92.7 | 74.5 |
| Artificial Analysis Coding Index | 25.6 | 6.7 |
| critpt | 0.6 | 0 |
| gdpval | 25.4 | 13.8 |
| GPQA Diamond | 81.4 | 64.1 |
| HLE | 16.5 | 4.6 |
| IFBench | 68.7 | 36.6 |
| LiveCodeBench | 84.5 | 51.3 |
| MathVista | 84.3 | 79.5 |
| mmlu_prox | 69.3 | 65 |
| MMMU-Pro | 69.2 | 57 |
| OmniScience Accuracy | 24.8 | 12.2 |
| OmniScience Non-Hallucination | 19.5 | 8.2 |
| scicode | 46.5 | 17.1 |
| Terminal-Bench Hard | 15.2 | 1.5 |
| τ²-Bench Telecom (AA run) | 55.6 | 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.