MiniMax H3 NVFP4 needs about 13.2 GB of memory at its main file (a 11.4 GB file) with an 8K context window, so it fits on a 16 GB GPU. This repository has 4 quants, from 11.4 GB to 18.7 GB.
MiniMax H3 NVFP4 GGUF diffusion model for NVIDIA Blackwell GPUs
MiniMax H3 NVFP4 is a diffusion‑model GGUF file set optimized for stable‑diffusion.cpp on Blackwell NVIDIA GPUs. It requires a Qwen3‑VL text encoder and VAE files for video and audio generation. The model is part of the MiniMax H3 family and is not a mixture of experts.
Good at
- Multimodal text‑to‑video generation
- Native stereo audio support
- High‑resolution up to 2K
Limits
- Requires Blackwell NVIDIA GPU
- Needs external text encoder and VAE files
- No standalone inference without stable‑diffusion.cpp
Best for
- Generating short video clips with audio
- Creating 2K video from 768p base
- Research on multimodal generation
Who it’s for. Developers and researchers who need a diffusion model for text‑to‑video tasks on NVIDIA GPUs. The model is suitable for those who can integrate it with stable‑diffusion.cpp and provide the necessary text encoder and VAE assets.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 11.4 GB | 13.2 GB | Too big | Get file → | |
| — | 11.4 GB | 13.2 GB | Too big | Get file → | |
| — | 18.7 GB | 21.4 GB | Too big | Get file → | |
| — | 18.7 GB | 21.4 GB | Too big | Get file → |
For 8‑GB VRAM, the 11.38‑GB GGUF files are too large; they require at least 13.2‑GB memory at 8K context. The 18.69‑GB files need about 21.4‑GB memory. Use a GPU with at least 14‑GB VRAM and enable "--max‑vram 14" with "--stream‑layers" to run the larger variants. No smaller quantizations are available in this set.
What GPU do you need for MiniMax H3 NVFP4?
These figures add the file size, an fp16 KV cache at an 8K context window and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: the best quant that fits is the main file (11.4 GB file, about 13.2 GB in use).
- 24 GB VRAM: the best quant that fits is the main file (18.7 GB file, about 21.4 GB in use).
Which MiniMax H3 NVFP4 quant should you choose?
If memory is tight, the smallest file is the first one in the table at 11.4 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about MiniMax H3 NVFP4
How much VRAM does MiniMax H3 NVFP4 need?
About 13.2 GB at its main file with an 8K context window. That is the 11.4 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run MiniMax H3 NVFP4 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 13.2 GB, so you would need a larger GPU or CPU offloading.
What is the file size of MiniMax H3 NVFP4 GGUF?
It depends on the quant: file is 11.4 GB, file is 11.4 GB, file is 18.7 GB, file is 18.7 GB.
Where can I download MiniMax H3 NVFP4 GGUF?
The files are on Hugging Face in the brurpo/MiniMax-H3-NVFP4-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does MiniMax H3 NVFP4 use?
Hugging Face lists the license as other. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run MiniMax H3 NVFP4 on an 8 GB VRAM GPU?
No. The smallest GGUF file requires 13.2 GB of VRAM at 8K context, so an 8 GB GPU cannot load it.
What additional files are needed for inference?
A Qwen3‑VL text encoder, a video VAE safetensor, and an audio VAE safetensor are required alongside the GGUF diffusion weights.
Is the model compatible with non‑NVIDIA GPUs?
The GGUF files are native NVFP4 tensors and are designed for Blackwell NVIDIA GPUs; other backends may load them but are not the target of this release.