Nemotron 3.5 Super VL 120B-A12B
- Made by
- NVIDIA
- Released
- Announced in
- Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt Modelsblogs.nvidia.com
- Event
- Nemotron 3.5 Super VL and Bolt
- API record
- /api/v1/models/nemotron-3-5-super-vl-120b-a12b
Benchmark results
| Benchmark | Score | Source check |
|---|---|---|
| Humanity's Last ExamNo tools · text-only subset; GPT-5.6 Luna judge | 31.79% | |
| in Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AI | 31.84% | CorroboratedThe same number is printed in 1 other lab document: 1 |
| Humanity's Last ExamWith tools · text-only subset; Python and web search; separate internal harness | 46.01% | 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. |
| SWE-bench ProThe source doesn't say whether this is the public or the commercial set | 51.78% | |
| Claude Code harness; 731-task public set; 2 repeats; git history removed | 49.45% | CorroboratedThe same number is printed in 1 other lab document: 1 |
| Mini-SWE-agent harness; 731-task public set; 2 repeats; git history removed | 47.40% | 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. |
| Codex harness; 731-task public set; 2 repeats; git history removed | 47.26% | 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. |
| Hermes harness; 731-task public set; 2 repeats; git history removed | 47.20% | 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. |
| Pi harness; 731-task public set; 2 repeats; git history removed | 46.92% | 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. |
| OpenCode harness; 731-task public set; 2 repeats; git history removed | 41.04% | 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. |
| GPQA Diamond | 89.52% | |
| in Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AI | 90.78% | 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. |
| BrowseComp | 78% | CorroboratedThe same number is printed in 1 other lab document: 1 |
| NeMo Gym harness with Exa search; discard-all context management · in Nemotron 3.5 Super and Lightning: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Models for Agentic AI | 78.91% | CorroboratedThe same number is printed in 1 other lab document: 1 |
| OpenCode harness · in Nemotron 3.5 Super and Lightning: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Models for Agentic AI | 72.35% | 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. |
| Exa web search · in Nemotron 3.5 Bolt 74B-A9B 24GB: A Hybrid Vision-Language Model for Local AI | 71.50% | 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. |
| Terminal-Bench 2.1 | 60.67% | 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. |
| Codex harness; v0.144.6; 89-task set; 3 repeats | 61.80% | 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. |
| Hermes harness; v0.18.2; 89-task set; 3 repeats | 61.42% | 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. |
| Claude Code harness; v2.1.126; 89-task set; 3 repeats | 59.55% | 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. |
| OpenCode harness; v1.17.8; 89-task set; 3 repeats | 55.06% | 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. |
| Mini-SWE-agent harness; v2.4.5; 89-task set; 3 repeats | 54.68% | 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. |
| Pi harness; v0.80.10; 89-task set; 3 repeats | 48.31% | 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-3.5-Super-VL-120B-A12B-NVFP4 model card (launch-day revision)huggingface.co · 15 Oct 2026
Original image 
Our capture of the source - 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 - Nemotron 3.5 Super and Lightning: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Models for Agentic AIresearch.nvidia.com · 15 Oct 2026
Our capture of page 23
