Across 23 shared benchmarks, Llama 3.1 Instruct 405B scores higher on 3 and Qwen3.5 122B A10B on 20. The widest gap is τ²-Bench Telecom (AA run), where Qwen3.5 122B A10B scores 93.6 against 19.
| Benchmark | Llama 3.1 Instruct 405B | Qwen3.5 122B A10B |
|---|---|---|
| AA Agentic Index | 6.3 | 21.3 |
| AA Intelligence | 8.5 | 32.8 |
| AA-LCR | 24.7 | 70.3 |
| AA-Omniscience | -17.1 | -41.5 |
| Artificial Analysis Coding Index | 14.5 | 45.7 |
| C-Eval | 72.5 | 91.9 |
| critpt | 0 | 0.9 |
| gdpval | 0 | 24.3 |
| GPQA Diamond | 51.5 | 86.6 |
| GSM8K | 96.8 | 94.5 |
| HLE | 4.2 | 47.5 |
| IFBench | 39 | 76.1 |
| ifeval | 88.6 | 93.4 |
| longbench_v2 | 36.1 | 60.2 |
| mmlu_redux | 86.2 | 94 |
| MMLU-Pro | 73.4 | 86.7 |
| mmmlu | 73.8 | 86.7 |
| OmniScience Accuracy | 23.2 | 24.4 |
| OmniScience Non-Hallucination | 49 | 12.9 |
| scicode | 29.9 | 42 |
| SWE-bench Verified | 24.5 | 72 |
| Terminal-Bench Hard | 6.8 | 31.1 |
| τ²-Bench Telecom (AA run) | 19 | 93.6 |
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.