| Benchmark | Gemini 3.5 Flash | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 70.4 | 14.4 |
| AA Intelligence | 52 | 7.7 |
| AA-LCR | 81 | 23 |
| AA-Omniscience | 21.2 | -68.4 |
| Artificial Analysis Coding Index | 70.1 | 6.7 |
| critpt | 13.1 | 0 |
| gdpval | 57.8 | 13.8 |
| GPQA Diamond | 92.2 | 64.1 |
| HLE | 42.7 | 4.6 |
| IFBench | 76.3 | 36.6 |
| MMMU-Pro | 84.3 | 57 |
| OmniScience Accuracy | 51.4 | 12.2 |
| OmniScience Non-Hallucination | 38.2 | 8.2 |
| OSWorld-Verified | 78.4 | 31.4 |
| scicode | 53.1 | 17.1 |
| Terminal-Bench Hard | 46.2 | 1.5 |
| τ²-Bench Telecom (AA run) | 95.6 | 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.