| Benchmark | Llama 3.1 Instruct 405B | Qwen3.6 Plus |
|---|---|---|
| AA Agentic Index | 6.3 | 29 |
| AA Intelligence | 8.5 | 40.5 |
| AA-LCR | 24.7 | 72.3 |
| AA-Omniscience | -17.1 | 2.6 |
| Artificial Analysis Coding Index | 14.5 | 54.5 |
| C-Eval | 72.5 | 93.3 |
| critpt | 0 | 2.9 |
| gdpval | 0 | 32 |
| GPQA Diamond | 51.5 | 90.4 |
| HLE | 4.2 | 28.8 |
| IFBench | 39 | 75.2 |
| ifeval | 88.6 | 94.3 |
| longbench_v2 | 36.1 | 62 |
| mmlu_redux | 86.2 | 94.5 |
| MMLU-Pro | 73.4 | 88.5 |
| mmmlu | 73.8 | 89.5 |
| OmniScience Accuracy | 23.2 | 26.4 |
| OmniScience Non-Hallucination | 49 | 68 |
| scicode | 29.9 | 40.7 |
| SWE-bench Pro | 11.2 | 56.6 |
| SWE-bench Verified | 24.5 | 78.8 |
| Terminal-Bench Hard | 6.8 | 43.9 |
| τ²-Bench Telecom (AA run) | 19 | 97.7 |
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.