MiniMax H3 curve needs about 17.6 GB of memory at Q5_1 (a 15.2 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 6 quants, from 11.5 GB (Q4_0) to 21.5 GB (Q8_0).
MiniMax H3 curve – 20.1B parameter GGUF model for multimodal generation
MiniMax H3 curve is a 20.1B parameter model in the Minimax family. It is available in GGUF format with Q4_0, Q5_1, and Q8_0 quantizations. The model supports text, image, video, and audio inputs and can generate video with stereo audio up to 2K resolution.
Good at
- Multimodal text‑to‑video generation
- Reference‑conditioned video output
- High‑resolution 2K video upscaling
Limits
- No vision support in llama.cpp
- Requires ComfyUI 0.30.0 or newer for full functionality
- License is other, not specified
Best for
- Creating short video clips from text prompts
- Generating reference‑conditioned video with audio
- Upscaling 768p video to 2K
Who it’s for. Developers and researchers building multimodal generation pipelines who need a compact 20.1B parameter model. It is suitable for users with access to ComfyUI and sufficient GPU memory for the chosen quantization.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_0 | 11.5 GB | 13.3 GB | Too big | Get file → | |
| Q4_0 | 11.5 GB | 13.3 GB | Too big | Get file → | |
| Q5_1 | 15.2 GB | 17.6 GB | Too big | Get file → | |
| Q5_1 | 15.2 GB | 17.6 GB | Too big | Get file → | |
| Q8_0 | 21.5 GB | 24.6 GB | Too big | Get file → | |
| Q8_0 | 21.5 GB | 24.6 GB | Too big | Get file → |
For 8 GB VRAM use Q4_0 (file size 11.46 GB, estimated memory 13.3 GB). For 16 GB VRAM use Q5_1 (file size 15.23 GB, estimated memory 17.6 GB). For 24 GB VRAM use Q8_0 (file size 21.5 GB, estimated memory 24.6 GB). Estimates are approximate.
What GPU do you need for MiniMax H3 curve?
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 Q4_0 (11.5 GB file, about 13.3 GB in use).
- 24 GB VRAM: the best quant that fits is Q5_1 (15.2 GB file, about 17.6 GB in use).
- 32 GB VRAM: the best quant that fits is Q8_0 (21.5 GB file, about 24.6 GB in use).
Which MiniMax H3 curve quant should you choose?
Q5_1 (15.2 GB) is the middle option in this repository. Closer to the original model than 4-bit, for a moderate increase in size. A good pick when you have memory to spare.
If memory is tight, the smallest file is Q4_0 at 11.5 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about MiniMax H3 curve
How much VRAM does MiniMax H3 curve need?
About 17.6 GB at Q5_1 with an 8K context window. That is the 15.2 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 curve on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 13.3 GB, so you would need a larger GPU or CPU offloading.
What is the file size of MiniMax H3 curve GGUF?
It depends on the quant: Q4_0 is 11.5 GB, Q4_0 is 11.5 GB, Q5_1 is 15.2 GB, Q5_1 is 15.2 GB, Q8_0 is 21.5 GB, Q8_0 is 21.5 GB.
Where can I download MiniMax H3 curve GGUF?
The files are on Hugging Face in the joeygambino/MiniMax-H3-curve-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 curve 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 curve on 8 GB VRAM?
Use the Q4_0 quantization; it requires about 13.3 GB of VRAM at 8k context.
Does the model support vision input in llama.cpp?
No, the model does not support vision in llama.cpp; it requires ComfyUI for multimodal processing.
What is the difference between the curve and original forms?
The curve form is about 40% smaller due to pruning and factorization, making the Q8_0 quantization fit a 24 GB card.