| Benchmark | Claude Sonnet 4 | Nemotron 3 Super 120B A12B |
|---|---|---|
| AA Agentic Index | 39.2 | 8.8 |
| AA Intelligence | 29.8 | 25.7 |
| AA-LCR | 69.7 | 60.3 |
| AA-Omniscience | 0.2 | -41.5 |
| AIME 2025 | 74 | 90.2 |
| Artificial Analysis Coding Index | 37.6 | 37.7 |
| browsecomp | 12.2 | 31.3 |
| critpt | 1.1 | 3.1 |
| gdpval | 31.2 | 9.9 |
| GPQA Diamond | 78 | 82.7 |
| HLE | 10.7 | 22.8 |
| HMMT 2025 | 15.9 | 94.7 |
| IFBench | 55 | 72.6 |
| LiveCodeBench | 68.5 | 82.1 |
| mmlu | 91.5 | 86 |
| MMLU-Pro | 84 | 83.8 |
| multichallenge | 57.1 | 55.2 |
| OmniScience Accuracy | 22.7 | 24.3 |
| OmniScience Non-Hallucination | 70.9 | 13 |
| scicode | 40 | 42 |
| SWE-bench Multilingual | 56.9 | 45.8 |
| SWE-bench Verified | 80.2 | 53.7 |
| Tau2 airline | 55.5 | 56.3 |
| Tau2 retail | 75 | 62.8 |
| TauBench V3 - Banking | 16.7 | 10.3 |
| Terminal-Bench | 36.4 | 25.8 |
| Terminal-Bench 2.1 | 36.3 | 38.6 |
| Terminal-Bench Hard | 31.1 | 28.8 |
| τ²-Bench Telecom (AA run) | 64.6 | 67.8 |
| τ³-Bench | 13.8 | 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.