Minimax-M3 needs about 266.4 GB of memory at Q4_K_M (a 264.9 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 22 quants, from 129.3 GB (IQ1_M) to 852.9 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.0 GB.
MiniMax-M3: 428 Billion‑parameter multimodal model with 1M token context
MiniMax-M3 is a native multimodal model with 1 M context and 428 B parameters. It uses MiniMax Sparse Attention for efficient long‑context processing. The model supports text, image, and video inputs and offers advanced reasoning modes.
Good at
- Long‑horizon agentic tasks
- Coding and coworking
- Multimodal reasoning
Limits
- Requires large VRAM for full precision
- Sparse attention not yet supported in llama.cpp
- License is other, not standard
Best for
- Developers building multimodal assistants
- Researchers testing long‑context performance
- Teams needing coding and cowork support
Who it’s for. Engineers and researchers who need a large multimodal model with million‑token context. The model is suitable for advanced AI applications that require deep semantic fusion across text, image, and video.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ1_M | 129.3 GB (4 parts) | 130.8 GB | Too big | Get file → | |
| IQ2_XXS | 134.9 GB (4 parts) | 136.4 GB | Too big | Get file → | |
| IQ2_M | 135.1 GB (4 parts) | 136.6 GB | Too big | Get file → | |
| Q2_K_XL | 143.9 GB (4 parts) | 145.4 GB | Too big | Get file → | |
| IQ3_XXS | 160.3 GB (5 parts) | 161.8 GB | Too big | Get file → | |
| IQ3_S | 175.6 GB (5 parts) | 177.2 GB | Too big | Get file → | |
| Q3_K_M | 195.5 GB (5 parts) | 197.0 GB | Too big | Get file → | |
| Q3_K_XL | 195.8 GB (5 parts) | 197.3 GB | Too big | Get file → | |
| IQ4_XS | 208.5 GB (6 parts) | 210.0 GB | Too big | Get file → | |
| IQ4_NL | 212.7 GB (6 parts) | 214.2 GB | Too big | Get file → | |
| Q4_K_S | 248.6 GB (7 parts) | 250.1 GB | Too big | Get file → | |
| MXFP4_MOE | 257.4 GB (7 parts) | 258.9 GB | Too big | Get file → | |
| Q4_K_M | 264.9 GB (7 parts) | 266.4 GB | Too big | Get file → | |
| Q4_K_XL | 265.8 GB (7 parts) | 267.3 GB | Too big | Get file → | |
| Q5_K_S | 299.7 GB (8 parts) | 301.2 GB | Too big | Get file → | |
| Q5_K_M | 318.7 GB (8 parts) | 320.2 GB | Too big | Get file → | |
| Q5_K_XL | 319.4 GB (8 parts) | 320.9 GB | Too big | Get file → | |
| Q6_K | 355.3 GB (9 parts) | 356.8 GB | Too big | Get file → | |
| Q6_K_XL | 388.1 GB (9 parts) | 389.6 GB | Too big | Get file → | |
| Q8_0 | 453.6 GB (11 parts) | 455.1 GB | Too big | Get file → | |
| Q8_K_XL | 465.2 GB (11 parts) | 466.7 GB | Too big | Get file → | |
| BF16 | 852.9 GB (18 parts) | 854.4 GB | Too big | Get file → |
For 8 GB VRAM, use IQ1_M (129 GB file, ~131 GB memory at 8 k context). For 16 GB VRAM, IQ2_XXS (134.9 GB file, ~136 GB memory). For 24 GB VRAM, Q4_K_M (264.93 GB file, ~266 GB memory). These estimates are approximate.
What GPU do you need for Minimax-M3?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) 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: no quant fits at an 8K context.
- 24 GB VRAM: no quant fits at an 8K context.
- 32 GB VRAM: no quant fits at an 8K context.
- 48 GB VRAM: no quant fits at an 8K context.
Which Minimax-M3 quant should you choose?
Q4_K_M is the usual starting point for Minimax-M3: it balances file size and quality well. Here it is a 264.9 GB file.
The Q8_0 file is 453.6 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is IQ1_M at 129.3 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Minimax-M3
How much VRAM does Minimax-M3 need?
About 266.4 GB at Q4_K_M with an 8K context window. That is the 264.9 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-M3 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 130.8 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Minimax-M3 GGUF?
It depends on the quant: IQ1_M is 129.3 GB, IQ2_XXS is 134.9 GB, IQ2_M is 135.1 GB, Q2_K_XL is 143.9 GB, IQ3_XXS is 160.3 GB, IQ3_S is 175.6 GB and 16 more.
What context length does Minimax-M3 support?
The GGUF header lists a trained context length of 1,048,576 tokens (1024K). The KV cache grows with the context you use: about 1.0 GB at 8K and 4.0 GB at 32K.
Where can I download Minimax-M3 GGUF?
The files are on Hugging Face in the unsloth/MiniMax-M3-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-M3 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-M3 on 8 GB VRAM?
Use the IQ1_M quant, which requires about 131 GB of memory at 8 k context, so 8 GB VRAM is insufficient for full inference.
Does MiniMax-M3 support vision in llama.cpp?
Vision is supported in the model, but llama.cpp currently lacks MiniMax Sparse Attention, so inference falls back to dense attention.
What reasoning modes does MiniMax-M3 offer?
It offers enabled, adaptive, and disabled reasoning modes via the `thinking` parameter for complex or latency‑sensitive tasks.