| Benchmark | Nemotron 3 Super 120B A12B | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 8.8 | 14.4 |
| AA Intelligence | 25.7 | 7.7 |
| AA-LCR | 60.3 | 23 |
| AA-Omniscience | -41.5 | -68.4 |
| AIME 2025 | 90.2 | 74.5 |
| Arena-Hard-V2 | 73.9 | 36.8 |
| Artificial Analysis Coding Index | 37.7 | 6.7 |
| critpt | 3.1 | 0 |
| gdpval | 9.9 | 13.8 |
| GPQA Diamond | 82.7 | 64.1 |
| HLE | 22.8 | 4.6 |
| HMMT 2025 | 94.7 | 53.1 |
| IFBench | 72.6 | 36.6 |
| LiveCodeBench | 82.1 | 51.3 |
| mmlu_prox | 79.4 | 65 |
| MMLU-Pro | 83.8 | 73.6 |
| OmniScience Accuracy | 24.3 | 12.2 |
| OmniScience Non-Hallucination | 13 | 8.2 |
| scicode | 42 | 17.1 |
| Terminal-Bench Hard | 28.8 | 1.5 |
| τ²-Bench Telecom (AA run) | 67.8 | 15.5 |
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.