Nemotron 3 family: which size to run
3 sizes, 30B–550B parameters · NVIDIA
Sizes
| Model | Architecture | |||||||
|---|---|---|---|---|---|---|---|---|
| Nemotron 3 Nano 30B A3B | 30B | MoE · 3.5B active | 262,144 | No LiveBench result for this model in release 2026-06-25 | 38* | 32* | 24.6 | 17.8 |
| Nemotron 3 Super 120B A12B | 120B | MoE · 12B active | 262,144 | No LiveBench result for this model in release 2026-06-25 | 98 | 92 | 82.5† | 68 |
| Nemotron 3 Ultra 550B A55B | 550B | MoE · 55B active | 262,144 | 67.4 | 421* | 415* | 359.2† | 347 |
Each size is its own model, with its own scores and memory needs. Q4 estimates are planning figures, not measurements; file sizes are published. The Mac or PC column is for a 32 GB or larger machine (8 GB reserve); machines under 32 GB reserve 25% of their memory (2 to 8 GB), so they need less. The publisher lists Nano 30B-A3B, Super 120B-A12B, Ultra 550B-A55B. * The model's page explains why its estimate differs.
† Unsloth's dynamic UD-Q4_K_M file, shown when the repository has no plain Q4_K_M file (the same rule as the model pages); the cell's title names the file.
How to read this table
Q4 means compressed to about 4 bits per weight. GGUF (llama.cpp, Ollama, LM Studio) and MLX (Apple silicon) are file formats; the sizes shown are the published 4-bit files. A dash means no value is published or collected for that size; the reason is in the cell's accessible text.
Which size fits my machine
The largest size whose Q4 planning estimate fits what each machine offers. A planning estimate, not a measurement and not a speed. Smaller machines are left out; the largest one that fits no size is kept to show the limit.
| Machine | Largest Nemotron 3 size with enough memory | ||
|---|---|---|---|
| Mac mini M4 · 32 GB unified | 32 | No size of this family has a Q4 estimate within what this machine offers | No size fits, so there is no estimate to compare |
| Mac mini M5 Pro · 64 GB unified | 56 | Nemotron 3 Nano 30B A3B | 38 |
| Mac Studio M5 Max · 128 GB unified | 104 | Nemotron 3 Super 120B A12B | 98 |
| Mac Studio · M5 Ultra · 36-core CPU / 80-core GPU · 512 GB unified | 392 | Nemotron 3 Super 120B A12B | 98 |
| RTX 4090 · 24 GB VRAM | 24 | No size of this family has a Q4 estimate within what this machine offers | No size fits, so there is no estimate to compare |
| RTX 5090 · 32 GB VRAM | 32 | Nemotron 3 Nano 30B A3B | 32 |
| DGX H100 · 8 × 80 GB | 640 | Nemotron 3 Ultra 550B A55B | 429 |
A dash means no size of this family fits that machine. The requirement is the size's Q4 estimate with that machine's own reserve: a graphics card reserves 2 GB, as in the sizes table; a Mac or PC reserves 25% of its memory (2 to 8 GB), so below 32 GB its figure is lower than the sizes table's. A server profile reserves 2 GB per graphics card, so its figure is higher.
Smallest machine for each size
The profile with the least memory that still has enough for the size's Q4 estimate, from the hardware catalog.
| Size | Smallest Mac or PC-memory profile | Smallest graphics-card profile | Server-class profile |
|---|---|---|---|
| Nemotron 3 Nano 30B A3B | Mac mini 2024 · M4 Pro · 48 GB | RTX 5090 · 32 GB | DGX H100 · 8 × 80 GB · 640 GB |
| Nemotron 3 Super 120B A12B | DGX Spark · 128 GB | No graphics-card profile in the catalog has enough memory for this size | DGX H100 · 8 × 80 GB · 640 GB |
| Nemotron 3 Ultra 550B A55B | No Mac, laptop or PC profile in the catalog has enough memory for this size | No graphics-card profile in the catalog has enough memory for this size | DGX H100 · 8 × 80 GB · 640 GB |
A dash means no profile of that kind in the catalog has enough memory for the size.
Memory by file precision
The same planning formula at three precisions (0.625, 1.125 and 2 bytes per parameter, plus working room and the system reserve). Estimates, not measurements.
| Size | ||||||
|---|---|---|---|---|---|---|
| Nemotron 3 Nano 30B A3B | 38* | 32* | 49* | 43* | 80* | 74* |
| Nemotron 3 Super 120B A12B | 98 | 92 | 170 | 164 | 296 | 290 |
| Nemotron 3 Ultra 550B A55B | 421* | 415* | 751* | 745* | 1,328* | 1,322* |
* The model's page explains why its estimate differs.
Family tree
Family, sizes and, for each size, the two 4-bit files the tables use (GGUF Q4_K_M and MLX 4-bit) when one was collected. Other precisions are not drawn here.
The same tree as a list:
- Nemotron 3 · NVIDIA
- Nano 30B A3B · 3.5B active
- GGUF Q4_K_M · 24.6 GB · unsloth/Nemotron-3-Nano-30B-A3B-GGUF, read Oct 23, 2026
- MLX 4-bit · 17.8 GB · lmstudio-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-4bit, read Oct 17, 2026
- Super 120B A12B · 12B active
- GGUF UD-Q4_K_M · 82.5 GB · unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF, read Oct 23, 2026
- MLX 4-bit · 68 GB · mlx-community/NVIDIA-Nemotron-3-Super-120B-A12B-4bit, read Oct 23, 2026
- Ultra 550B A55B · 55B active
- GGUF UD-Q4_K_M · 359.2 GB · unsloth/NVIDIA-Nemotron-3-Ultra-550B-A55B-GGUF, read Oct 23, 2026
- MLX 4-bit · 347 GB · mlx-community/Nemotron-3-Ultra-550B-A55B-4bit, read Oct 23, 2026
- Nano 30B A3B · 3.5B active
Learn more
See Nemotron 3 Nano 30B A3B
The smallest size of the family: its scores, memory estimate and the machines it runs on.
Sources
Where the family and its sizes are named, and where each number comes from.
- nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 model card · “Nemotron-3-Nano-30B-A3B-BF16 is a large language model (LLM) trained from scratch by NVIDIA ... The model has 3.5B active parameters and 30B parameters in total.” · read Oct 17, 2026
- nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 model card · “Like other models in the family, it responds to user queries ... Distinct from the Nano model, the Super model incorporates Multi-Token Prediction (MTP) layers” · read Oct 17, 2026
- nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 model card · “Like the Super model, the Ultra model incorporates ...” · read Oct 17, 2026
- Nemotron 3 Nano 30B A3B model card
- Nemotron 3 Super 120B A12B model card
- Nemotron 3 Ultra 550B A55B model card
- LiveBench 2026-06-25 · the only quality source on this page; a dash means the release does not include the model · snapshot collected Sep 26, 2026
- Published file sizes: Hugging Face repositories, read Oct 23, 2026
- Methodology · how the Q4 planning estimate is computed · Learn