| Benchmark | DeepSeek-V4-Flash | Nemotron 3 Super 120B A12B |
|---|---|---|
| AA Agentic Index | 61.3 | 8.8 |
| AA Intelligence | 51.8 | 25.7 |
| AA-LCR | 74.3 | 60.3 |
| AA-Omniscience | -14.3 | -41.5 |
| aime_2026 | 95.8 | 91.7 |
| Artificial Analysis Coding Index | 69.1 | 37.7 |
| browsecomp | 73.2 | 31.3 |
| critpt | 16.6 | 3.1 |
| gdpval | 52.9 | 9.9 |
| GPQA Diamond | 90.8 | 82.7 |
| HLE | 38.6 | 22.8 |
| hmmt_feb_2026 | 94.8 | 85.2 |
| IFBench | 79.2 | 72.6 |
| itbenchSre | 31.5 | 1.1 |
| LiveCodeBench | 91.6 | 82.1 |
| MMLU-Pro | 86.4 | 83.8 |
| OmniScience Accuracy | 40.4 | 24.3 |
| OmniScience Non-Hallucination | 15.6 | 13 |
| scicode | 49.9 | 42 |
| SWE-bench Multilingual | 73.3 | 45.8 |
| SWE-bench Verified | 79 | 53.7 |
| TauBench V3 - Banking | 39.4 | 10.3 |
| Terminal-Bench 2.0 | 56.9 | 31 |
| Terminal-Bench 2.1 | 82.7 | 38.6 |
| Terminal-Bench Hard | 38.6 | 28.8 |
| τ²-Bench Telecom (AA run) | 95.6 | 67.8 |
| τ³-Bench | 22.9 | 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.