| Benchmark | GPT-5 | Nemotron 3 Super 120B A12B |
|---|---|---|
| AA Agentic Index | 49.7 | 8.8 |
| AA Intelligence | 35.3 | 25.7 |
| AA-LCR | 76.3 | 60.3 |
| AA-Omniscience | -8.7 | -41.5 |
| AIME 2025 | 95 | 90.2 |
| AIME25 no tools | 94.6 | 92.2 |
| Artificial Analysis Coding Index | 38.9 | 37.7 |
| browsecomp | 54.9 | 31.3 |
| critpt | 5.7 | 3.1 |
| gdpval | 32.5 | 9.9 |
| GPQA Diamond | 87.3 | 82.7 |
| GSM8K | -1.8 | 90.7 |
| HLE | 28.5 | 22.8 |
| HMMT 2025 | 93.3 | 94.7 |
| humaneval | 93.4 | 80.5 |
| IFBench | 73.1 | 72.6 |
| LiveCodeBench | 85 | 82.1 |
| MMLU-Pro | 87.5 | 83.8 |
| multichallenge | 71.1 | 55.2 |
| OmniScience Accuracy | 40.3 | 24.3 |
| OmniScience Non-Hallucination | 21 | 13 |
| scicode | 43 | 42 |
| SWE-bench Multilingual | 55.3 | 45.8 |
| SWE-bench Verified | 74.9 | 53.7 |
| Tau2 airline | 62.6 | 56.3 |
| Tau2 retail | 81.1 | 62.8 |
| TauBench V3 - Banking | 22.1 | 10.3 |
| Terminal-Bench | 43.8 | 25.8 |
| Terminal-Bench 2.0 | 35.2 | 31 |
| Terminal-Bench 2.1 | 35.2 | 38.6 |
| Terminal-Bench Hard | 37.9 | 28.8 |
| τ²-Bench Telecom (AA run) | 86.5 | 67.8 |
| τ³-Bench | 19.6 | 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.