| Benchmark | Nemotron 3 Super 120B A12B | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 25.7 | 10 |
| AA-LCR | 60.3 | 20.3 |
| AA-Omniscience | -41.5 | -52.2 |
| arc_challenge | 96.1 | 94.5 |
| Artificial Analysis Coding Index | 37.7 | 11.9 |
| critpt | 3.1 | 0 |
| GPQA Diamond | 82.7 | 49.1 |
| GSM8K | 90.7 | 95.8 |
| hellaswag | 89 | 84.8 |
| HLE | 22.8 | 4.2 |
| humaneval | 80.5 | 86.6 |
| IFBench | 72.6 | 36.9 |
| LiveCodeBench | 82.1 | 55.5 |
| MBPP | 81.7 | 72.6 |
| mmlu | 86 | 86.1 |
| MMLU-Pro | 83.8 | 71.6 |
| OmniScience Accuracy | 24.3 | 17.6 |
| OmniScience Non-Hallucination | 13 | 15.3 |
| OpenBookQA | 48.6 | 96.2 |
| scicode | 42 | 26.7 |
| SWE-bench Verified | 53.7 | 23.8 |
| Terminal-Bench Hard | 28.8 | 4.5 |
| winogrande | 78.9 | 82.3 |
| τ²-Bench Telecom (AA run) | 67.8 | 34.5 |
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.