| Benchmark | Llama 3.1 Instruct 405B | Qwen3.5 397B A17B |
|---|---|---|
| AA Agentic Index | 6.3 | 53.3 |
| AA Intelligence | 8.5 | 34.3 |
| AA-LCR | 24.7 | 72.7 |
| AA-Omniscience | -17.1 | -30.8 |
| Artificial Analysis Coding Index | 14.5 | 48.2 |
| C-Eval | 72.5 | 93 |
| critpt | 0 | 1.7 |
| gdpval | 0 | 35.8 |
| GPQA Diamond | 51.5 | 89.3 |
| GSM8K | 96.8 | 66.7 |
| HLE | 4.2 | 29 |
| IFBench | 39 | 78.8 |
| ifeval | 88.6 | 92.6 |
| longbench_v2 | 36.1 | 63.2 |
| mmlu_redux | 86.2 | 94.9 |
| MMLU-Pro | 73.4 | 87.8 |
| mmmlu | 73.8 | 88.5 |
| OmniScience Accuracy | 23.2 | 30.8 |
| OmniScience Non-Hallucination | 49 | 17.3 |
| scicode | 29.9 | 42 |
| SWE-bench Pro | 11.2 | 50.9 |
| SWE-bench Verified | 24.5 | 76.4 |
| Terminal-Bench Hard | 6.8 | 40.9 |
| τ²-Bench Telecom (AA run) | 19 | 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.