MiniMax M2 THRIFT needs about 107.2 GB of memory at Q4_K_M (a 104.6 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 26 quants, from 0.4 GB to 183.4 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 8.3 GB.
MiniMax M2 THRIFT – a pruned, efficient LLM for coding and reasoning
MiniMax M2 THRIFT is a 173 billion‑parameter model pruned to 25 % of the base. It supports 196 608‑token context and is optimized for local inference. The model is released under the MIT license and is suitable for chat, coding, and reasoning tasks.
Good at
- Coding assistance
- Reasoning tasks
- Tool‑use integration
Limits
- Large context requires high VRAM
- No vision capabilities
- Requires llama.cpp for inference
Best for
- Local chatbots with tight VRAM budgets
- Batch code generation
- Edge deployments needing fast responses
Who it’s for. Developers and researchers who need a lightweight version of MiniMax M2 for local or air‑gapped environments. The model is ideal for teams that require fast inference and lower memory usage while maintaining coding and reasoning performance.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.4 GB | 3.0 GB | Fits | Get file → | |
| IQ1_S | 35.7 GB | 38.3 GB | Too big | Get file → | |
| IQ1_M | 37.2 GB | 39.8 GB | Too big | Get file → | |
| IQ2_XXS | 41.5 GB | 44.1 GB | Too big | Get file → | |
| IQ2_XS | 47.8 GB | 50.4 GB | Too big | Get file → | |
| IQ2_S | 48.0 GB | 50.6 GB | Too big | Get file → | |
| IQ2_M | 54.5 GB (2 parts) | 57.1 GB | Too big | Get file → | |
| Q2_K | 60.4 GB (2 parts) | 63.0 GB | Too big | Get file → | |
| Q2_K_L | 61.0 GB (2 parts) | 63.6 GB | Too big | Get file → | |
| IQ3_XXS | 68.1 GB (2 parts) | 70.7 GB | Too big | Get file → | |
| IQ3_XS | 70.9 GB (2 parts) | 73.5 GB | Too big | Get file → | |
| Q3_K_S | 74.9 GB (2 parts) | 77.5 GB | Too big | Get file → | |
| IQ3_M | 78.6 GB (2 parts) | 81.1 GB | Too big | Get file → | |
| Q3_K_M | 78.6 GB (2 parts) | 81.1 GB | Too big | Get file → | |
| Q3_K_L | 81.7 GB (3 parts) | 84.3 GB | Too big | Get file → | |
| Q3_K_XL | 82.2 GB (3 parts) | 84.8 GB | Too big | Get file → | |
| IQ4_XS | 92.3 GB (3 parts) | 94.9 GB | Too big | Get file → | |
| IQ4_NL | 97.6 GB (3 parts) | 100.2 GB | Too big | Get file → | |
| Q4_0 | 99.1 GB (3 parts) | 101.7 GB | Too big | Get file → | |
| Q4_K_S | 101.0 GB (3 parts) | 103.6 GB | Too big | Get file → | |
| Q4_K_M | 104.6 GB (3 parts) | 107.2 GB | Too big | Get file → | |
| Q4_1 | 108.2 GB (3 parts) | 110.7 GB | Too big | Get file → | |
| Q5_K_S | 118.9 GB (3 parts) | 121.5 GB | Too big | Get file → | |
| Q5_K_M | 122.5 GB (4 parts) | 125.1 GB | Too big | Get file → | |
| Q6_K | 141.7 GB (4 parts) | 144.3 GB | Too big | Get file → | |
| Q8_0 | 183.4 GB (5 parts) | 186.0 GB | Too big | Get file → |
For 8 GB VRAM, use IQ1_S (0.37 GB file, 3 GB memory at 8 k context). For 16 GB VRAM, IQ2_XS (47.82 GB file, 50.4 GB memory) or Q2_K (60.42 GB file, 63 GB memory). For 24 GB VRAM, Q3_K_M (78.57 GB file, 81.1 GB memory) or Q4_K_M (104.61 GB file, 107.2 GB memory). Estimates are approximate.
What GPU do you need for MiniMax M2 THRIFT?
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: the best quant that fits is the main file (0.4 GB file, about 3.0 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (0.4 GB file, about 3.0 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (0.4 GB file, about 3.0 GB in use).
- 12 GB VRAM: the best quant that fits is the main file (0.4 GB file, about 3.0 GB in use).
- 16 GB VRAM: the best quant that fits is the main file (0.4 GB file, about 3.0 GB in use).
- 24 GB VRAM: the best quant that fits is the main file (0.4 GB file, about 3.0 GB in use).
- 32 GB VRAM: the best quant that fits is the main file (0.4 GB file, about 3.0 GB in use).
- 48 GB VRAM: the best quant that fits is IQ2_XXS (41.5 GB file, about 44.1 GB in use).
Which MiniMax M2 THRIFT quant should you choose?
Q4_K_M is the usual starting point for MiniMax M2 THRIFT: it balances file size and quality well. Here it is a 104.6 GB file.
The Q8_0 file is 183.4 GB, about 1.8 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 the first one in the table at 0.4 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 M2 THRIFT
How much VRAM does MiniMax M2 THRIFT need?
About 107.2 GB at Q4_K_M with an 8K context window. That is the 104.6 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 M2 THRIFT on an 8 GB GPU?
Yes, with the smaller files: the 0.4 GB file needs about 3.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of MiniMax M2 THRIFT GGUF?
It depends on the quant: file is 0.4 GB, IQ1_S is 35.7 GB, IQ1_M is 37.2 GB, IQ2_XXS is 41.5 GB, IQ2_XS is 47.8 GB, IQ2_S is 48.0 GB and 20 more.
What context length does MiniMax M2 THRIFT support?
The GGUF header lists a trained context length of 196,608 tokens (192K). The KV cache grows with the context you use: about 2.1 GB at 8K and 8.3 GB at 32K.
Where can I download MiniMax M2 THRIFT GGUF?
The files are on Hugging Face in the bartowski/VibeStudio_MiniMax-M2-THRIFT-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 M2 THRIFT use?
Hugging Face lists the license as mit. 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 M2 THRIFT on 8 GB VRAM?
Yes, the IQ1_S quant uses 0.37 GB file size and requires about 3 GB memory at 8 k context, fitting within 8 GB VRAM.
Does the model support vision tasks?
No, the model has no vision capabilities.
What license does the model use?
The model is released under the MIT license.