| Benchmark | Llama 3.1 Instruct 405B | Qwen3.5 35B A3B |
|---|---|---|
| AA Agentic Index | 6.3 | 44.1 |
| AA Intelligence | 8.5 | 29.9 |
| AA-LCR | 24.7 | 68.3 |
| AA-Omniscience | -17.1 | -48.1 |
| arc_challenge | 96.9 | 95.4 |
| Artificial Analysis Coding Index | 14.5 | 37 |
| C-Eval | 72.5 | 90.2 |
| critpt | 0 | 0.9 |
| gdpval | 0 | 20.3 |
| GPQA Diamond | 51.5 | 84.5 |
| GSM8K | 96.8 | 90.1 |
| hellaswag | 89.2 | 85.6 |
| HLE | 4.2 | 47.4 |
| humaneval | 89 | 66.5 |
| IFBench | 39 | 72.5 |
| ifeval | 88.6 | 91.9 |
| longbench_v2 | 36.1 | 59 |
| MBPP | 73.4 | 70.8 |
| mmlu | 88.6 | 81.1 |
| mmlu_redux | 86.2 | 93.3 |
| MMLU-Pro | 73.4 | 85.3 |
| mmmlu | 73.8 | 85.2 |
| OmniScience Accuracy | 23.2 | 20.1 |
| OmniScience Non-Hallucination | 49 | 14.6 |
| OpenBookQA | 94 | 44.2 |
| scicode | 29.9 | 37.7 |
| SWE-bench Pro | 11.2 | 44.6 |
| SWE-bench Verified | 24.5 | 70 |
| Terminal-Bench Hard | 6.8 | 26.5 |
| winogrande | 86.7 | 79.2 |
| τ²-Bench Telecom (AA run) | 19 | 89.2 |
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.