Nemotron 3 Super 120B A12B
NVIDIA · nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4
Q4 file 82.5 GB (UD-Q4_K_M) · runs well from 128 GB · Weights license: Publisher terms: NVIDIA Nemotron Open Model License
Quality
No result in LiveBench 2026-06-25, LMArena 2026-10-22 Not in LiveBench 2026-06-25 or LMArena 2026-10-22
Reported by the publisher
- MMLU-Pro
- 83.33General Knowledge
- HMMT Feb25 (with tools)
- 95.36Reasoning
- GPQA (no tools)
- 79.42Reasoning
- LiveCodeBench (v6 2024-08↔2025-05)
- 78.44Reasoning
- LiveCodeBench (v5 2024-07↔2024-12)
- 80.56Reasoning
- SciCode (subtask)
- 40.83Reasoning
- HLE (no tools)
- 17.42Reasoning
- Terminal Bench (hard subset)
- 24.48Agentic
Every reported result
- Airline
- 54.75TauBench V2
- Retail
- 63.38TauBench V2
- Telecom
- 63.27TauBench V2
- Average
- 60.46TauBench V2
- IFBench (prompt)
- 73.30Chat & Instruction Following
- Scale AI Multi-Challenge
- 52.8Chat & Instruction Following
- Arena-Hard-V2 (Hard Prompt)
- 76.00Chat & Instruction Following
- AA-LCR
- 58.06Long Context
- RULER-500 @ 128k (500 samples per task)
- 95.99Long Context
- RULER-500 @ 256k (500 samples per task)
- 96.52Long Context
- RULER-500 @ 512k (500 samples per task)
- 96.23Long Context
- MMLU-ProX (avg over languages)
- 79.37Multilingual
Across Nemotron 3 sizes
| Benchmark (publisher settings) | Nano 30B A3B | Super 120B A12B | Ultra 550B A55B |
|---|---|---|---|
| IFBench (prompt) | This size's card does not report this result under these settings | 73.30 | 82.3 |
| AA-LCR | This size's card does not report this result under these settings | 58.06 | 65.5 |
Values as printed on the card, with the publisher's own settings (card section: License/Terms of Use › Benchmarks); column “Nemotron-3-Super NVFP4”. Not comparable across publishers and not used in any ranking. Model card
Run it
Run it
File UD-Q4_K_M · 82.5 GB · 3 parts · unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF · What is Q4?
Measured 15–15 tok/s · UD-Q4_K_XL · 1 machine · 2 sources · all rows
llama-server -hf unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF:UD-Q4_K_M- Install llama.cpp
- Run the command above
- Ask at localhost:8080:
Explain what a context window is, in two sentences.
Runs on
Fast memory holds it 14
The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.
Show all 14
One memory tier short 1
The same machines with the next memory size up would hold it in fast memory.
Nemotron 3 Super 120B A12B is a text model from NVIDIA with 120B parameters, 12B of them active per token (mixture of experts). It was released on March 11, 2026 and reads up to 262,144 tokens of context. Its Q4 weights plus working room come to about 90 GB before the machine's reserve; the smallest reference machine whose fast memory holds that is Mac Studio M5 Max · 128 GB unified (~98 GB with its reserve). We collected 25 downloadable GGUF and MLX files from 2 publisher sources; the most downloaded GGUF source is unsloth's (212.7K downloads on the Hub). It has no result in LiveBench 2026-06-25, so no quality score is shown.
Hardware profiles grouped by the Q4 planning estimate on each machine (dedicated GPUs reserve 2 GB; unified and system memory 25%, 2 to 8 GB): the fast memory holds it, it is within memory but outside the fast part (layers offloaded to PC RAM, or above the macOS GPU share), or the machine is one memory tier short. Each card opens its hardware profile. Not a measured fit or speed.
On the reference machines
| Where Q4 sits | ||||
|---|---|---|---|---|
| Mac mini M2 · 8 GB unifiedLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~92 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| Mac mini M4 · 16 GB unifiedLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~94 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| Mac mini M4 · 24 GB unifiedLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~96 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| Mac mini M4 · 32 GB unifiedLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~98 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| Mac mini M5 Pro · 64 GB unifiedLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~98 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| Mac Studio M5 Max · 128 GB unifiedFast memory holds itUD-IQ4_XS262,144 tokens | ~98 GB | Fast memory holds it | UD-IQ4_XS | 262,144 tokens |
| Mac Studio · M5 Ultra · 36-core CPU / 80-core GPU · 512 GB unifiedFast memory holds itQ8_0262,144 tokens | ~98 GB | Fast memory holds it | Q8_0 | 262,144 tokens |
| RTX 4060 · 8 GB VRAMLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~92 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| RTX 3060 · 12 GB VRAMLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~92 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| RTX 5080 · 16 GB VRAMLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~92 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| RTX 4090 · 24 GB VRAMLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~92 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| RTX 5090 · 32 GB VRAMLarger than its memoryNo file on the site's quant ladder fits the fast memory at 8,192 tokensNo file in the previous column | ~92 GB | Larger than its memory | No file on the site's quant ladder fits the fast memory at 8,192 tokens | No file in the previous column |
| DGX H100 · 8 × 80 GBFast memory holds itQ8_0262,144 tokens | ~106 GB | Fast memory holds it | Q8_0 | 262,144 tokens |
The file column uses the site's quant ladder (Q8_0 to Q3_K_M, MLX 8 to 4-bit on a Mac): the largest collected file whose size × 1.2 plus the context cache and the machine's reserve fits the fast memory. Planning estimates, not measured runs.
Compare Nemotron 3 Super 120B A12B using its publisher card. No overall score appears in the LiveBench 2026-06-25 snapshot.
The Q4 weights plus working room come to about 90 GB from 120B stated parameters, before the machine's reserve (Runs on adds it per machine). It is not a measured minimum; context, batch size, runtime and quantization change actual memory use.
Versions to download
Files you can download for this checkpoint. The section opens on the most downloaded collected GGUF source, unsloth (212.7K Hub downloads), at UD-Q4_K_M. On a Mac it opens on the most downloaded MLX source. Sizes are the Hub's published bytes. What is Q4?
unsloth · 212.7K downloads
unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF · UD-Q4_K_M/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_M-00001-of-00003.gguf · 82.5 GB
Run this file
llama-server -hf unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF:UD-Q4_K_Mlms get https://huggingface.co/unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF- Ollama: Ollama's documented hf.co form covers single-file GGUFs; this file has 3 parts
- mlx-lm: mlx-lm runs MLX repositories; pick the MLX version on a Mac
Quality scores on this page belong to the original checkpoint; these files are not scored or ranked. File sizes read 2026-10-23 · download counts read 2026-10-17 · commands: Hugging Face Hub docs: Use Ollama with any GGUF model · llama.cpp server README (-hf <user>/<model>[:quant]) · LM Studio docs: lms get with a Hugging Face URL · mlx-lm README (mlx_lm.generate --model)
Fine-tunes, adapters and other versions
- Quantized versions
- 57
- Fine-tunes
- 17
- Adapters · merges
- 152 · 0
Browse them on the Hugging Face model tree · read 2026-10-23
Uncensored and abliterated versions
Some community uploads remove the model's safety refusals. Quality may drop, the original publisher does not support them, claims about them are the uploader's, and the original license still applies. This site never recommends or ranks them.
0 derivative repositories have “uncensored” or “abliterated” in their name (read 2026-10-23).
Measured speed
Tokens per second by machine
Output tok/s · 2 machines · every row in the table · What is a token?
| DGX Spark | 19.94 ± 0.16 | Q4_K (reported as 'Q4_K' by llama-bench; GGUF from ggml-org) | llama.cpp CUDA | ggml-org/Nemotron-3-Super-120B-GGUF · Q4_K (65.10 GiB); not the NVFP4 checkpoint linked on the model page |
|---|---|---|---|---|
| Framework Desktop 128GB | 14.86 ± 0.06 | UD-Q4_K_XL (llama-bench: Q4_K - Medium, 78.02 GiB) | llama.cpp Vulkan (Mesa RADV, toolbox vulkan_radv) | NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_XL-00001-of-00003.gguf (Unsloth UD-Q4_K_XL, 78.02 GiB) |
| Framework Desktop 128GB | 14.85 ± 0.04 | UD-Q4_K_XL (llama-bench: Q4_K - Medium, 78.02 GiB) | llama.cpp ROCm 7.2.3 (toolbox rocm-7.2.3) | NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_XL-00001-of-00003.gguf (Unsloth UD-Q4_K_XL, 78.02 GiB) |
Sources: llama.cpp benches/nemotron · DGX Spark · kyuz0/amd-strix-halo-toolboxes raw llama-bench log (GitHub) · community benchmark · kyuz0/amd-strix-halo-toolboxes raw llama-bench log (GitHub) · community benchmark · not estimated
How to run
The memory needed above is the Q4 load before a machine's reserve. Formula: parameters × 0.625 bytes, plus 20% working allowance; each machine then adds its system reserve: 25% of unified or system memory, 2 to 8 GB, or 2 GB per GPU on dedicated cards. This is neither measured VRAM nor checkpoint download size.
Q4 planning estimate only. Check checkpoint format, runtime support, KV cache, context, batch size and offload before choosing hardware.
Evidence level: Generic estimate only; no execution reproduced by this site.
- Publisher sampling defaults
- temperature 1 · top_p 0.95generation_config.json · read 2026-10-23source
- Chat template markers
- tool calls · thinkingmarkers found in the template, not a test of the feature · read 2026-10-23source
Before downloading
- Check the exact checkpoint and license linked here.
- Confirm a compatible runtime and precision or quantization for your platform.
- Allow for context, batch size and any extra media components beyond the weights.
Technical details
What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.
- Checkpoint / version
- nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4
- Task and modality
- language and reasoning · text
- Count and basis
- 120B · manufacturer specification · 12B active
- Context
- 262,144 tokens published for this model
- Architecture
- MoE
- Released
- nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 model card (Hugging Face) · read 2026-10-09source
- Runtime and precision
- No local execution verified.
- Languages
- en, fr, es, it, de, ja, zhModel card metadata · read 2026-10-25source
- Commercial use
- License text not read: nvidia.com was unreachable on 2026-10-18 and the repositories ship no copy
Every stored field
- Revision checked
- ff433f5493e25d631c9f12b5d55c674229923d02
- Architecture
- NemotronHForCausalLM
- Library
- transformers
- Hub task tag
- text-generation
- Formats
- safetensors
- Access
- Open download
- Publisher kind
- Official
- Repository created
- 2026-03-10creation is not the model's release
- Repository updated
- 2026-08-24a commit is not a release date
- Hub downloads (30 days) · likes
- 535,065 · 455this repository only · read 2026-10-25
- Read from the Hub page
- transformers class, task tags, license nameHub overview and file listing · read 2026-10-18
- Catalog state
- Hub config or weights found; no inventory benchmark link
Sources
Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.
- Publisher checkpoint and model card
- Weight license and use terms
- Quality benchmark source · LiveBench 2026-06-25 overall
Catalog reviewed: 2026-09-30. Benchmark release: 2026-06-25; snapshot collected: 2026-09-26.










