DE

Deepseek-V3.1

DeepSeekGGUF by unslothUpdated 2025-09-22
Download on Hugging Face
Parameters670B
Context160K
Layers61
Architecturedeepseek2
Smallest quant170.5 GB
Quick answer

Deepseek-V3.1 needs about 407.0 GB of memory at Q4_K_M (a 405.4 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 27 quants, from 170.5 GB (TQ1_0) to 1342.3 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.

Deepseek‑V3.1 – 670 Billion‑parameter hybrid LLM with 163 K token context

Deepseek‑V3.1 is a 670 Billion‑parameter model that supports both thinking and non‑thinking modes. It extends the 32K and 128K context phases to 630 B and 209 B tokens respectively. The model is released under the MIT license and is available in multiple GGUF quantisations.

Good at

  • Hybrid thinking mode
  • Improved tool‑calling
  • Long‑context handling

Limits

  • Large VRAM requirement for higher quants
  • No vision support
  • Requires llama.cpp with --jinja

Best for

  • Chatbots needing long‑context reasoning
  • Agent tasks with tool calls
  • Research on hybrid LLM behaviour

Who it’s for. Developers and researchers who need a large, hybrid LLM that can switch between thinking and non‑thinking modes. The model is suitable for building chat applications and agent systems that require extended context and tool‑calling capabilities.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
TQ1_0170.5 GB 172.1 GB
Too big Get file →
IQ1_S192.3 GB (4 parts) 193.9 GB
Too big Get file →
IQ1_M206.8 GB (5 parts) 208.4 GB
Too big Get file →
IQ2_XXS225.9 GB (5 parts) 227.5 GB
Too big Get file →
IQ2_M235.5 GB (5 parts) 237.1 GB
Too big Get file →
Q2_K245.7 GB (5 parts) 247.3 GB
Too big Get file →
Q2_K_L245.9 GB (5 parts) 247.5 GB
Too big Get file →
Q2_K_XL255.7 GB (6 parts) 257.3 GB
Too big Get file →
IQ3_XXS279.7 GB (6 parts) 281.3 GB
Too big Get file →
Q3_K_S290.4 GB (6 parts) 292.0 GB
Too big Get file →
Q3_K_XL300.0 GB (7 parts) 301.6 GB
Too big Get file →
Q3_K_M320.5 GB (7 parts) 322.0 GB
Too big Get file →
IQ4_XS358.3 GB (8 parts) 359.9 GB
Too big Get file →
IQ4_NL379.1 GB (8 parts) 380.7 GB
Too big Get file →
Q4_0380.1 GB (8 parts) 381.7 GB
Too big Get file →
Q4_K_S381.1 GB (8 parts) 382.7 GB
Too big Get file →
Q4_K_XL386.9 GB (8 parts) 388.5 GB
Too big Get file →
Q4_K_M405.4 GB (9 parts) 407.0 GB
Too big Get file →
Q4_1420.8 GB (9 parts) 422.4 GB
Too big Get file →
Q5_K_S462.6 GB (10 parts) 464.2 GB
Too big Get file →
Q5_K_M476.2 GB (10 parts) 477.7 GB
Too big Get file →
Q5_K_XL484.6 GB (10 parts) 486.2 GB
Too big Get file →
Q6_K551.3 GB (12 parts) 552.9 GB
Too big Get file →
Q6_K_XL574.5 GB (12 parts) 576.1 GB
Too big Get file →
Q8_0713.3 GB (15 parts) 714.9 GB
Too big Get file →
Q8_K_XL780.6 GB (17 parts) 782.2 GB
Too big Get file →
BF161342.3 GB (30 parts) 1343.9 GB
Too big Get file →

For 8 GB VRAM, use Q4_0 (380 GB file, ~382 GB memory). For 16 GB VRAM, Q4_K_M (405 GB file, ~407 GB memory) is a good balance. For 24 GB VRAM, Q5_K_M (476 GB file, ~478 GB memory) provides higher quality while staying within memory limits. Sharding is required for all quants with split_into_shards >1.

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: 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 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 405.4 GB file.

The Q8_0 file is 713.3 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 TQ1_0 at 170.5 GB, with the quality trade-offs that come with fewer bits. Lower bit counts mean smaller files and more quality loss.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturedeepseek2
Parameters670B (approximate, mixture of experts)
Layers61
Trained context163,840 tokens
KV cache (fp16)130 KB per token

Source

Quantized byunsloth
FamilyDeepSeek
Licensemit
Downloads (HF)13,102
Last updated2025-09-22

Frequently asked questions about Deepseek-V3.1

How much VRAM does Deepseek-V3.1 need?

About 407.0 GB at Q4_K_M with an 8K context window. That is the 405.4 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?

Not at an 8K context. The smallest file needs about 172.1 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Deepseek-V3.1 GGUF?

It depends on the quant: TQ1_0 is 170.5 GB, IQ1_S is 192.3 GB, IQ1_M is 206.8 GB, IQ2_XXS is 225.9 GB, IQ2_M is 235.5 GB, Q2_K is 245.7 GB and 21 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 unsloth/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?

Use the Q4_0 quant; it requires approximately 382 GB of VRAM at 8 k context, which is not feasible on 8 GB. A smaller quant such as Q4_0 is still too large for 8 GB VRAM.

Does Deepseek‑V3.1 support vision?

No, the model does not include vision capabilities.

What chat template is required for llama.cpp?

Use the --jinja flag with the provided chat template to enable hybrid thinking and non‑thinking modes.

Other GGUFs of DeepSeek-V3.1

Similar models

Browse all →

More from unsloth

All by unsloth →

More DeepSeek models

All DeepSeek →

Other large models

All large models →