| Benchmark | o1 | Qwen3.5 397B A17B |
|---|---|---|
| AA Agentic Index | 31.1 | 53.3 |
| AA Intelligence | 24 | 34.3 |
| AA-LCR | 63.3 | 72.7 |
| AA-Omniscience | -10.6 | -30.8 |
| arena_vision | 1193 | 1249 |
| Artificial Analysis Coding Index | 39.7 | 48.2 |
| critpt | 0.3 | 1.7 |
| gdpval | 11.5 | 35.8 |
| GPQA Diamond | 75.7 | 89.3 |
| GSM8K | 97.1 | 66.7 |
| HLE | 7.7 | 29 |
| IFBench | 70.3 | 78.8 |
| mmmlu | 87.7 | 88.5 |
| MMMU | 77.6 | 85 |
| OmniScience Accuracy | 34.7 | 30.8 |
| OmniScience Non-Hallucination | 30.7 | 17.3 |
| scicode | 35.8 | 42 |
| strongreject | 97 | 99.4 |
| SWE-bench Verified | 48.9 | 76.4 |
| Terminal-Bench Hard | 12.9 | 40.9 |
| τ²-Bench Telecom (AA run) | 62.6 | 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.