Mistral-Small-3.2-24B-Instruct-2506 needs about 16.2 GB of memory at Q4_K_M (a 14.3 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 29 quants, from 6.5 GB (IQ1_S) to 47.2 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 5.4 GB.
Mistral‑Small‑3.2‑24B‑Instruct‑2506 – 24 billion‑parameter instruction model with vision and 128k context
Mistral‑Small‑3.2‑24B‑Instruct‑2506 is a 24 billion‑parameter instruction model built on the Mistral architecture. It supports vision tasks and a 128k token context window. The model is released under Apache‑2.0 and is available in multiple GGUF quantizations.
Good at
- Vision analysis
- Long‑context instruction following
- Multilingual text generation
Limits
- No dedicated vision library support in llama.cpp
- Requires large VRAM for high‑bit quantizations
- No built‑in tokenizer for non‑English languages
Best for
- Developing chatbots that need image understanding
- Generating long documents with context up to 128k tokens
- Building multilingual assistants
Who it’s for. Developers and researchers who need a versatile instruction model with vision capabilities and a large context window. The model is suitable for both commercial and non‑commercial use under the Apache‑2.0 license.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ1_S | 6.5 GB | 8.4 GB | Too big | Get file → | |
| IQ1_M | 7.0 GB | 8.9 GB | Too big | Get file → | |
| IQ2_XXS | 7.4 GB | 9.2 GB | Too big | Get file → | |
| IQ2_XS | 8.0 GB | 9.8 GB | Too big | Get file → | |
| IQ2_S | 8.2 GB | 10.1 GB | Too big | Get file → | |
| Q2_K_S | 8.4 GB | 10.2 GB | Too big | Get file → | |
| IQ2_M | 8.7 GB | 10.5 GB | Too big | Get file → | |
| Q2_K_M | 9.1 GB | 11.0 GB | Too big | Get file → | |
| IQ3_XXS | 9.5 GB | 11.4 GB | Too big | Get file → | |
| IQ3_XS | 10.0 GB | 11.8 GB | Too big | Get file → | |
| IQ3_S | 10.5 GB | 12.3 GB | Too big | Get file → | |
| IQ3_M | 10.6 GB | 12.5 GB | Too big | Get file → | |
| Q3_K_S | 10.6 GB | 12.5 GB | Too big | Get file → | |
| Q3_K_M | 11.6 GB | 13.5 GB | Too big | Get file → | |
| IQ4_XS | 12.8 GB | 14.6 GB | Too big | Get file → | |
| Q4_0 | 13.3 GB | 15.1 GB | Too big | Get file → | |
| IQ4_NL | 13.5 GB | 15.3 GB | Too big | Get file → | |
| Q4_K_S | 13.8 GB | 15.7 GB | Too big | Get file → | |
| Q4_K_M | 14.3 GB | 16.2 GB | Too big | Get file → | |
| Q4_1 | 14.7 GB | 16.6 GB | Too big | Get file → | |
| Q5_0 | 16.2 GB | 18.1 GB | Too big | Get file → | |
| Q5_K_S | 16.6 GB | 18.4 GB | Too big | Get file → | |
| Q5_K_M | 16.9 GB | 18.7 GB | Too big | Get file → | |
| Q5_1 | 17.7 GB | 19.5 GB | Too big | Get file → | |
| Q6_K_M | 19.3 GB | 21.2 GB | Too big | Get file → | |
| Q8_0 | 25.1 GB | 26.9 GB | Too big | Get file → | |
| BF16 | 33.6 GB | 35.4 GB | Too big | Get file → | |
| F16 | 33.6 GB | 35.4 GB | Too big | Get file → | |
| BF16 | 47.2 GB | 49.0 GB | Too big | Get file → |
For 8 GB VRAM, use IQ1_S (6.52 GB, 8.4 GB memory at 8k context) or IQ1_M (7.05 GB, 8.9 GB memory). For 16 GB VRAM, Q4_K_M (14.34 GB, 16.2 GB memory) or Q5_K_M (16.88 GB, 18.7 GB memory) provide a good balance of size and precision. Higher‑bit quantizations such as Q8_0 or F16 require 25 GB+ VRAM and are recommended only for high‑end GPUs.
What GPU do you need for Mistral-Small-3.2-24B-Instruct-2506?
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: the best quant that fits is IQ3_XS (10.0 GB file, about 11.8 GB in use).
- 16 GB VRAM: the best quant that fits is Q4_K_S (13.8 GB file, about 15.7 GB in use).
- 24 GB VRAM: the best quant that fits is Q6_K_M (19.3 GB file, about 21.2 GB in use).
- 32 GB VRAM: the best quant that fits is Q8_0 (25.1 GB file, about 26.9 GB in use).
- 48 GB VRAM: the best quant that fits is BF16 (33.6 GB file, about 35.4 GB in use).
Which Mistral-Small-3.2-24B-Instruct-2506 quant should you choose?
Q4_K_M is the usual starting point for Mistral-Small-3.2-24B-Instruct-2506: it balances file size and quality well. Here it is a 14.3 GB file.
The Q8_0 file is 25.1 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_S at 6.5 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
Frequently asked questions about Mistral-Small-3.2-24B-Instruct-2506
How much VRAM does Mistral-Small-3.2-24B-Instruct-2506 need?
About 16.2 GB at Q4_K_M with an 8K context window. That is the 14.3 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 Mistral-Small-3.2-24B-Instruct-2506 on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 8.4 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Mistral-Small-3.2-24B-Instruct-2506 GGUF?
It depends on the quant: IQ1_S is 6.5 GB, IQ1_M is 7.0 GB, IQ2_XXS is 7.4 GB, IQ2_XS is 8.0 GB, IQ2_S is 8.2 GB, Q2_K_S is 8.4 GB and 23 more.
What context length does Mistral-Small-3.2-24B-Instruct-2506 support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 1.3 GB at 8K and 5.4 GB at 32K.
Where can I download Mistral-Small-3.2-24B-Instruct-2506 GGUF?
The files are on Hugging Face in the Mungert/Mistral-Small-3.2-24B-Instruct-2506-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 Mistral-Small-3.2-24B-Instruct-2506 use?
Hugging Face lists the license as apache-2.0. 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 Mistral‑Small‑3.2‑24B‑Instruct‑2506 on 8 GB VRAM?
Yes, using IQ1_S or IQ1_M quantizations. They fit within 8 GB VRAM and use approximately 8.4–8.9 GB memory at 8k context.
Does this model support vision tasks in llama.cpp?
Vision capabilities are present, but llama.cpp does not provide a dedicated vision library. Image processing must be handled externally.
What context length does the model support?
The model supports up to 128k tokens, with a trained context of 131072 tokens.