| Benchmark | DeepSeek-R1 | Nemotron 3 Super 120B A12B |
|---|---|---|
| AA Agentic Index | 3.1 | 8.8 |
| AA Intelligence | 18.6 | 25.7 |
| AA-LCR | 56 | 60.3 |
| AA-Omniscience | -31.3 | -41.5 |
| AIME 2025 | 87.5 | 90.2 |
| Artificial Analysis Coding Index | 24.6 | 37.7 |
| browsecomp | 8.9 | 31.3 |
| critpt | 0.6 | 3.1 |
| gdpval | 1.5 | 9.9 |
| GPQA Diamond | 81 | 82.7 |
| HLE | 17.7 | 22.8 |
| HMMT 2025 | 79.4 | 94.7 |
| IFBench | 39 | 72.6 |
| LiveCodeBench | 84.4 | 82.1 |
| mmlu | 92.9 | 86 |
| mmlu_prox | 75.5 | 79.4 |
| MMLU-Pro | 85 | 83.8 |
| multichallenge | 45 | 55.2 |
| OmniScience Accuracy | 30.7 | 24.3 |
| OmniScience Non-Hallucination | 10.5 | 13 |
| scicode | 35.7 | 42 |
| SWE-bench Multilingual | 30.5 | 45.8 |
| SWE-bench Verified | 57.6 | 53.7 |
| TauBench V3 - Banking | 6.4 | 10.3 |
| Terminal-Bench | 5.7 | 25.8 |
| Terminal-Bench 2.1 | 19.1 | 38.6 |
| Terminal-Bench Hard | 6.1 | 28.8 |
| τ²-Bench Telecom (AA run) | 11.4 | 67.8 |
| τ³-Bench | 6.4 | 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.