| Benchmark | Gemini 2.5 Flash | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 18.8 | 14.4 |
| AA Intelligence | 17.7 | 7.7 |
| AA-LCR | 65.7 | 23 |
| AA-Omniscience | -29.8 | -68.4 |
| AIME 2024 | 82.3 | 72.9 |
| AIME 2025 | 72 | 74.5 |
| Artificial Analysis Coding Index | 22.2 | 6.7 |
| critpt | 1.4 | 0 |
| gdpval | 11.9 | 13.8 |
| GPQA Diamond | 82.8 | 64.1 |
| HLE | 12.1 | 4.6 |
| IFBench | 50.3 | 36.6 |
| LiveCodeBench | 76.2 | 51.3 |
| MMMU-Pro | 69.1 | 57 |
| OmniScience Accuracy | 26.1 | 12.2 |
| OmniScience Non-Hallucination | 24.7 | 8.2 |
| scicode | 35.9 | 17.1 |
| Terminal-Bench Hard | 13.6 | 1.5 |
| τ²-Bench Telecom (AA run) | 31.6 | 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.