| Benchmark | DeepSeek-V3.2 | Nemotron 3 Super 120B A12B |
|---|---|---|
| AA Agentic Index | 39.8 | 8.8 |
| AA Intelligence | 32.8 | 25.7 |
| AA-LCR | 70.7 | 60.3 |
| AA-Omniscience | -22.5 | -41.5 |
| AIME 2025 | 93.1 | 90.2 |
| aime_2026 | 95.1 | 91.7 |
| AIME25 no tools | 89.3 | 92.2 |
| apexAgents | 14.5 | 1.8 |
| Artificial Analysis Coding Index | 44.2 | 37.7 |
| browsecomp | 51.4 | 31.3 |
| critpt | 2.9 | 3.1 |
| gdpval | 18.8 | 9.9 |
| GPQA Diamond | 84 | 82.7 |
| HLE | 40.8 | 22.8 |
| HMMT 2025 | 90.2 | 94.7 |
| hmmt_feb_2026 | 84.1 | 85.2 |
| IFBench | 61 | 72.6 |
| LiveCodeBench | 86 | 82.1 |
| MMLU-Pro | 86 | 83.8 |
| OmniScience Accuracy | 33 | 24.3 |
| OmniScience Non-Hallucination | 17.3 | 13 |
| scicode | 39 | 42 |
| SWE-bench Multilingual | 70.2 | 45.8 |
| SWE-bench Verified | 73.1 | 53.7 |
| TauBench V3 - Banking | 18.8 | 10.3 |
| Terminal-Bench | 37.7 | 25.8 |
| Terminal-Bench 2.0 | 46.4 | 31 |
| Terminal-Bench 2.1 | 46.8 | 38.6 |
| Terminal-Bench Hard | 35.6 | 28.8 |
| τ²-Bench Telecom (AA run) | 90.6 | 67.8 |
| τ³-Bench | 69.2 | 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.