| Benchmark | o3 | Qwen3.6 35B A3B |
|---|---|---|
| AA Agentic Index | 36.1 | 21.6 |
| AA Intelligence | 31.1 | 32.1 |
| AA-LCR | 73.3 | 66.7 |
| AA-Omniscience | -15.3 | -22.2 |
| Artificial Analysis Coding Index | 38.4 | 41.9 |
| browsecomp | 49.7 | 48.7 |
| critpt | 1.1 | 0.3 |
| gdpval | 12.8 | 27.8 |
| GPQA Diamond | 83.3 | 86 |
| HLE | 20.3 | 22.2 |
| hmmt_feb_2025 | 77.5 | 90.7 |
| IFBench | 71.4 | 64.4 |
| MathVista | 86.8 | 86.4 |
| MMLU-Pro | 85 | 85.6 |
| MMMU | 82.9 | 81.7 |
| MMMU-Pro | 76.4 | 75.3 |
| OmniScience Accuracy | 38.6 | 18.8 |
| OmniScience Non-Hallucination | 12.9 | 49.5 |
| scicode | 41 | 35.8 |
| SWE-bench Verified | 69.1 | 73.4 |
| Terminal-Bench Hard | 37.1 | 34.8 |
| VideoMMMU | 83.3 | 83.7 |
| τ²-Bench Telecom (AA run) | 80.7 | 95.3 |
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.