| Benchmark | GPT-4o | Qwen3.6 35B A3B |
|---|---|---|
| AA Agentic Index | 9.7 | 21.6 |
| AA Intelligence | 11.2 | 32.1 |
| AA-LCR | 55 | 66.7 |
| AA-Omniscience | -10.5 | -22.2 |
| AI2D | 94.2 | 92.7 |
| Artificial Analysis Coding Index | 24.2 | 41.9 |
| C-Eval | 76 | 90 |
| critpt | 0 | 0.3 |
| gdpval | 0 | 27.8 |
| GPQA Diamond | 70.1 | 86 |
| HLE | 5.3 | 22.2 |
| IFBench | 36 | 64.4 |
| LiveCodeBench v6 | 30.9 | 80.4 |
| MathVista | 63.8 | 86.4 |
| mmlu_redux | 88 | 93.3 |
| MMLU-Pro | 74.7 | 85.6 |
| MMMU | 72.2 | 81.7 |
| MMMU-Pro | 59.9 | 75.3 |
| OmniScience Accuracy | 23.7 | 18.8 |
| OmniScience Non-Hallucination | 62.1 | 49.5 |
| RealWorldQA | 75.4 | 85.3 |
| scicode | 33.4 | 35.8 |
| supergpqa | 42.4 | 64.7 |
| SWE-bench Verified | 38.8 | 73.4 |
| Terminal-Bench Hard | 8.3 | 34.8 |
| VideoMMMU | 61.2 | 83.7 |
| τ²-Bench Telecom (AA run) | 28.9 | 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.