| Benchmark | Gemini 2.5 Flash | Qwen3.5 397B A17B |
|---|---|---|
| AA Agentic Index | 18.8 | 53.3 |
| AA Intelligence | 17.7 | 34.3 |
| AA-LCR | 65.7 | 72.7 |
| AA-Omniscience | -29.8 | -30.8 |
| arena_text_factuality | 1437 | 1445 |
| arena_vision | 1214 | 1249 |
| Artificial Analysis Coding Index | 22.2 | 48.2 |
| critpt | 1.4 | 1.7 |
| gdpval | 11.9 | 35.8 |
| Global-MMLU-Lite | 88.4 | 90 |
| GPQA Diamond | 82.8 | 89.3 |
| HLE | 12.1 | 29 |
| hmmt_feb_2025 | 64.2 | 94.8 |
| IFBench | 50.3 | 78.8 |
| MMMU | 79.7 | 85 |
| MMMU-Pro | 69.1 | 79 |
| OmniScience Accuracy | 26.1 | 30.8 |
| OmniScience Non-Hallucination | 24.7 | 17.3 |
| scicode | 35.9 | 42 |
| SWE-bench Verified | 60.4 | 76.4 |
| Terminal-Bench Hard | 13.6 | 40.9 |
| τ²-Bench Telecom (AA run) | 31.6 | 95.6 |
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.