| Benchmark | Gemini 3 Flash Preview | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 49.7 | 14.4 |
| AA Intelligence | 38.7 | 7.7 |
| AA-LCR | 73 | 23 |
| AA-Omniscience | 10.1 | -68.4 |
| AIME 2025 | 99.7 | 74.5 |
| Artificial Analysis Coding Index | 42.6 | 6.7 |
| critpt | 8.6 | 0 |
| gdpval | 35.2 | 13.8 |
| GPQA Diamond | 90.4 | 64.1 |
| HLE | 43.5 | 4.6 |
| IFBench | 78 | 36.6 |
| MMMU-Pro | 81.2 | 57 |
| OmniScience Accuracy | 53.4 | 12.2 |
| OmniScience Non-Hallucination | 7.6 | 8.2 |
| OSWorld-Verified | 65.1 | 31.4 |
| scicode | 50.6 | 17.1 |
| screenspot_pro_no_tools | 69.1 | 49.2 |
| Terminal-Bench Hard | 38.6 | 1.5 |
| VideoMMMU | 86.9 | 69.4 |
| τ²-Bench Telecom (AA run) | 80.4 | 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.