Nemotron 3 Super 120B-A12B
- Made by
- NVIDIA
- Released
- Announced in
- New NVIDIA Nemotron 3 Super Delivers 5x Higher Throughput for Agentic AIblogs.nvidia.com
- Event
- Nemotron 3 Super
- API record
- /api/v1/models/nemotron-3-super-120b-a12b
Benchmark results
| Benchmark | Score | Source check |
|---|---|---|
| Humanity's Last ExamNo tools | 18.26% | |
| reported later, in Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AI | 20.30% | CorroboratedThe same number is printed in 1 other lab document: 1 |
| Humanity's Last ExamWith tools | 22.8% | |
| SWE-bench VerifiedOpenHands harness | 60.47% | |
| GPQA Diamondno tools | 79.23% | |
| reported later, in Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AI | 78.60% | CorroboratedThe same number is printed in 1 other lab document: 1 |
| BrowseCompinternal search harness with SerpAPI; no context management | 31.28% | |
| Exa web search · reported later, in Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AI | 22.77% | CorroboratedThe same number is printed in 1 other lab document: 1 |
| Terminal-Bench 2.0Harbor harness | 31.00% | |
| AIME 2025No tools | 90.21% | |
| Terminal-Bench 2.1reported later, in Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AI | 39.61% | CorroboratedThe same number is printed in 1 other lab document: 1 |
| MMLU-Pro | 83.73% | |
| HMMT February 2025No tools | 93.67% | |
| LiveCodeBenchLiveCodeBench v6, problems from Aug 2024 to May 2025 | 78.69% | |
| LiveCodeBench v5, problems from Jul to Dec 2024 | 81.19% | |
| τ²-benchTelecom · default prompt | 64.36% | |
| τ²-benchRetail · default prompt | 62.83% | |
| τ²-benchAirline · default prompt | 56.25% | |
| MCP Atlaspublic 500-task set; reasoning on · reported later, in Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AI | 52.8% | CheckedA person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
Sources
Reported by NVIDIA
- nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8 model card (launch-day revision)huggingface.co · 11 Mar 2026

Original image 
Our capture of the source - Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoningresearch.nvidia.com · 11 Mar 2026

Our capture of page 31 - Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AIresearch.nvidia.com · 15 Oct 2026

Our capture of page 16