| Benchmark | Claude Sonnet 4.5 | Nemotron 3 Super 120B A12B |
|---|---|---|
| AA Agentic Index | 50.6 | 8.8 |
| AA Intelligence | 37.4 | 25.7 |
| AA-LCR | 68.3 | 60.3 |
| AA-Omniscience | -0.1 | -41.5 |
| AIME 2025 | 88 | 90.2 |
| AIME25 no tools | 88 | 92.2 |
| Artificial Analysis Coding Index | 52.1 | 37.7 |
| browsecomp | 24.1 | 31.3 |
| critpt | 1.1 | 3.1 |
| gdpval | 40.4 | 9.9 |
| GPQA Diamond | 83.4 | 82.7 |
| HLE | 19.8 | 22.8 |
| HMMT 2025 | 74.6 | 94.7 |
| IFBench | 57.3 | 72.6 |
| LiveCodeBench | 71 | 82.1 |
| MMLU-Pro | 88.2 | 83.8 |
| multichallenge | 55.3 | 55.2 |
| OmniScience Accuracy | 32.9 | 24.3 |
| OmniScience Non-Hallucination | 50.9 | 13 |
| scicode | 45 | 42 |
| SWE-bench Multilingual | 68 | 45.8 |
| SWE-bench Verified | 82 | 53.7 |
| TauBench V3 - Banking | 24.5 | 10.3 |
| Terminal-Bench | 51 | 25.8 |
| Terminal-Bench 2.0 | 50 | 31 |
| Terminal-Bench 2.1 | 55.8 | 38.6 |
| Terminal-Bench Hard | 35.6 | 28.8 |
| τ²-Bench Telecom (AA run) | 78.1 | 67.8 |
| τ³-Bench | 19 | 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.