DeepSeek V3.1 needs about 410.8 GB of memory at Q4_K_M (a 409.2 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 24 quants, from 1.0 GB to 713.3 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.
DeepSeek V3.1 – a 670B hybrid LLM with long‑context support
DeepSeek V3.1 is a 670 billion‑parameter hybrid model that supports both thinking and non‑thinking modes. It offers extended context up to 163 840 tokens and improved tool‑calling performance. The model is released under the MIT license and is available in multiple GGUF quantizations.
Good at
- Long‑context reasoning
- Hybrid thinking mode
- Tool‑calling efficiency
Limits
- Requires large VRAM for full precision
- No vision support
- Limited to text only
Best for
- Chatbots needing long context
- Agent tasks with tool calls
- Research on hybrid reasoning models
Who it’s for. Developers building conversational agents that need extended context and tool integration. Researchers exploring hybrid thinking architectures may also benefit.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 1.0 GB | 2.6 GB | Fits | Get file → | |
| IQ1_M | 147.4 GB (4 parts) | 149.0 GB | Too big | Get file → | |
| IQ2_XXS | 164.1 GB (5 parts) | 165.6 GB | Too big | Get file → | |
| IQ2_XS | 188.4 GB (5 parts) | 190.0 GB | Too big | Get file → | |
| IQ2_S | 189.6 GB (5 parts) | 191.2 GB | Too big | Get file → | |
| IQ2_M | 215.0 GB (6 parts) | 216.6 GB | Too big | Get file → | |
| Q2_K | 237.8 GB (7 parts) | 239.4 GB | Too big | Get file → | |
| Q2_K_L | 238.7 GB (7 parts) | 240.3 GB | Too big | Get file → | |
| IQ3_XXS | 267.6 GB (7 parts) | 269.2 GB | Too big | Get file → | |
| IQ3_XS | 277.1 GB (8 parts) | 278.7 GB | Too big | Get file → | |
| Q3_K_S | 293.4 GB (8 parts) | 294.9 GB | Too big | Get file → | |
| IQ3_M | 307.9 GB (8 parts) | 309.5 GB | Too big | Get file → | |
| Q3_K_M | 307.9 GB (8 parts) | 309.5 GB | Too big | Get file → | |
| Q3_K_L | 319.7 GB (9 parts) | 321.3 GB | Too big | Get file → | |
| Q3_K_XL | 320.5 GB (9 parts) | 322.1 GB | Too big | Get file → | |
| IQ4_XS | 360.0 GB (10 parts) | 361.6 GB | Too big | Get file → | |
| IQ4_NL | 380.5 GB (10 parts) | 382.1 GB | Too big | Get file → | |
| Q4_0 | 386.4 GB (10 parts) | 388.0 GB | Too big | Get file → | |
| Q4_K_S | 394.2 GB (11 parts) | 395.7 GB | Too big | Get file → | |
| Q4_K_M | 409.2 GB (11 parts) | 410.8 GB | Too big | Get file → | |
| Q4_1 | 421.0 GB (11 parts) | 422.6 GB | Too big | Get file → | |
| Q5_K_M | 478.3 GB (13 parts) | 479.9 GB | Too big | Get file → | |
| Q6_K | 552.4 GB (15 parts) | 554.0 GB | Too big | Get file → | |
| Q8_0 | 713.3 GB (20 parts) | 714.9 GB | Too big | Get file → |
For 8 GB VRAM, use Q4_0 (≈388 GB memory at 8k context). For 16 GB VRAM, Q3_K_M (≈310 GB) or Q4_K_S (≈396 GB) are suitable. For 32 GB VRAM, Q2_K (≈240 GB) or Q3_K_L (≈321 GB) provide a balance of quality and memory.
What GPU do you need for DeepSeek V3.1?
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 (1.0 GB file, about 2.6 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (1.0 GB file, about 2.6 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (1.0 GB file, about 2.6 GB in use).
- 12 GB VRAM: the best quant that fits is the main file (1.0 GB file, about 2.6 GB in use).
- 16 GB VRAM: the best quant that fits is the main file (1.0 GB file, about 2.6 GB in use).
- 24 GB VRAM: the best quant that fits is the main file (1.0 GB file, about 2.6 GB in use).
- 32 GB VRAM: the best quant that fits is the main file (1.0 GB file, about 2.6 GB in use).
- 48 GB VRAM: the best quant that fits is the main file (1.0 GB file, about 2.6 GB in use).
Which DeepSeek V3.1 quant should you choose?
Q4_K_M is the usual starting point for DeepSeek V3.1: it balances file size and quality well. Here it is a 409.2 GB file.
The Q8_0 file is 713.3 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 the first one in the table at 1.0 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 DeepSeek V3.1
How much VRAM does DeepSeek V3.1 need?
About 410.8 GB at Q4_K_M with an 8K context window. That is the 409.2 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 DeepSeek V3.1 on an 8 GB GPU?
Yes, with the smaller files: the 1.0 GB file needs about 2.6 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of DeepSeek V3.1 GGUF?
It depends on the quant: file is 1.0 GB, IQ1_M is 147.4 GB, IQ2_XXS is 164.1 GB, IQ2_XS is 188.4 GB, IQ2_S is 189.6 GB, IQ2_M is 215.0 GB and 18 more.
What context length does DeepSeek V3.1 support?
The GGUF header lists a trained context length of 163,840 tokens (160K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.
Where can I download DeepSeek V3.1 GGUF?
The files are on Hugging Face in the bartowski/deepseek-ai_DeepSeek-V3.1-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 DeepSeek V3.1 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 DeepSeek V3.1 on 8 GB VRAM?
Yes, using the Q4_0 quantization which requires approximately 388 GB of VRAM at 8k context. It is the smallest quant that fits within 8 GB.
Does DeepSeek V3.1 support vision tasks?
No, the model is text‑only and does not include vision capabilities.