| Benchmark | DeepSeek-V4-Pro | Nemotron 3 Super 120B A12B |
|---|---|---|
| AA Agentic Index | 63.3 | 8.8 |
| AA Intelligence | 53 | 25.7 |
| AA-LCR | 70 | 60.3 |
| AA-Omniscience | -10.6 | -41.5 |
| aime_2026 | 96.7 | 91.7 |
| apexAgents | 24.3 | 1.8 |
| Artificial Analysis Coding Index | 59.4 | 37.7 |
| browsecomp | 83.4 | 31.3 |
| critpt | 13 | 3.1 |
| gdpval | 49 | 9.9 |
| GPQA Diamond | 90.5 | 82.7 |
| HLE | 48.2 | 22.8 |
| hmmt_feb_2026 | 95.2 | 85.2 |
| IFBench | 76.5 | 72.6 |
| itbenchSre | 38.3 | 1.1 |
| LiveCodeBench | 93.5 | 82.1 |
| MMLU-Pro | 87.5 | 83.8 |
| OmniScience Accuracy | 43 | 24.3 |
| OmniScience Non-Hallucination | 12.2 | 13 |
| scicode | 50 | 42 |
| SWE-bench Multilingual | 76.2 | 45.8 |
| SWE-bench Verified | 80.6 | 53.7 |
| TauBench V3 - Banking | 30.1 | 10.3 |
| Terminal-Bench 2.0 | 67.9 | 31 |
| Terminal-Bench 2.1 | 72.1 | 38.6 |
| Terminal-Bench Hard | 46.2 | 28.8 |
| τ²-Bench Telecom (AA run) | 96.2 | 67.8 |
| τ³-Bench | 25.8 | 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.