| Benchmark | DeepSeek-V3 | Nemotron 3 Super 120B A12B |
|---|---|---|
| AA Agentic Index | 1.6 | 8.8 |
| AA Intelligence | 15.4 | 25.7 |
| AA-LCR | 41.3 | 60.3 |
| AA-Omniscience | -37.6 | -41.5 |
| AIME 2025 | 51.3 | 90.2 |
| arc_challenge | 95.3 | 96.1 |
| Artificial Analysis Coding Index | 23 | 37.7 |
| critpt | 0 | 3.1 |
| gdpval | 0 | 9.9 |
| GPQA Diamond | 68.4 | 82.7 |
| GSM8K | 96.7 | 90.7 |
| hellaswag | 88.9 | 89 |
| HLE | 5.2 | 22.8 |
| HMMT 2025 | 27.5 | 94.7 |
| humaneval | 92.1 | 80.5 |
| IFBench | 41 | 72.6 |
| LiveCodeBench | 49.6 | 82.1 |
| MBPP | 75.4 | 81.7 |
| mmlu | 89.4 | 86 |
| mmlu_prox | 70.5 | 79.4 |
| MMLU-Pro | 81.2 | 83.8 |
| multichallenge | 31.4 | 55.2 |
| OmniScience Accuracy | 25.4 | 24.3 |
| OmniScience Non-Hallucination | 23.3 | 13 |
| OpenBookQA | 95.4 | 48.6 |
| scicode | 35.8 | 42 |
| SWE-bench Verified | 42 | 53.7 |
| Tau2 airline | 39 | 56.3 |
| Tau2 retail | 69.1 | 62.8 |
| TauBench V3 - Banking | 4.7 | 10.3 |
| Terminal-Bench 2.1 | 16.9 | 38.6 |
| Terminal-Bench Hard | 15.2 | 28.8 |
| winogrande | 84.9 | 78.9 |
| τ²-Bench Telecom (AA run) | 47.1 | 67.8 |
| τ³-Bench | 4.7 | 10.1 |
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.