| Benchmark | Llama 3.1 Instruct 405B | Qwen3.6 35B A3B |
|---|---|---|
| AA Agentic Index | 6.3 | 21.6 |
| AA Intelligence | 8.5 | 32.1 |
| AA-LCR | 24.7 | 66.7 |
| AA-Omniscience | -17.1 | -22.2 |
| Artificial Analysis Coding Index | 14.5 | 41.9 |
| C-Eval | 72.5 | 90 |
| critpt | 0 | 0.3 |
| gdpval | 0 | 27.8 |
| GPQA Diamond | 51.5 | 86 |
| HLE | 4.2 | 22.2 |
| IFBench | 39 | 64.4 |
| mmlu_redux | 86.2 | 93.3 |
| MMLU-Pro | 73.4 | 85.6 |
| OmniScience Accuracy | 23.2 | 18.8 |
| OmniScience Non-Hallucination | 49 | 49.5 |
| scicode | 29.9 | 35.8 |
| SWE-bench Pro | 11.2 | 49.5 |
| SWE-bench Verified | 24.5 | 73.4 |
| Terminal-Bench Hard | 6.8 | 34.8 |
| τ²-Bench Telecom (AA run) | 19 | 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.