| Benchmark | Llama 3.1 Instruct 405B | Nemotron 3 Super 120B A12B |
|---|---|---|
| AA Agentic Index | 6.3 | 8.8 |
| AA Intelligence | 8.5 | 25.7 |
| AA-LCR | 24.7 | 60.3 |
| AA-Omniscience | -17.1 | -41.5 |
| arc_challenge | 96.9 | 96.1 |
| Artificial Analysis Coding Index | 14.5 | 37.7 |
| critpt | 0 | 3.1 |
| gdpval | 0 | 9.9 |
| GPQA Diamond | 51.5 | 82.7 |
| GSM8K | 96.8 | 90.7 |
| hellaswag | 89.2 | 89 |
| HLE | 4.2 | 22.8 |
| humaneval | 89 | 80.5 |
| IFBench | 39 | 72.6 |
| LiveCodeBench | 30.1 | 82.1 |
| MBPP | 73.4 | 81.7 |
| mmlu | 88.6 | 86 |
| MMLU-Pro | 73.4 | 83.8 |
| OmniScience Accuracy | 23.2 | 24.3 |
| OmniScience Non-Hallucination | 49 | 13 |
| OpenBookQA | 94 | 48.6 |
| scicode | 29.9 | 42 |
| SWE-bench Verified | 24.5 | 53.7 |
| Terminal-Bench Hard | 6.8 | 28.8 |
| winogrande | 86.7 | 78.9 |
| τ²-Bench Telecom (AA run) | 19 | 67.8 |
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.