DeepSeek · Over 80B · Mixture of experts

DeepSeek V3.2

685.4 billion parameters in total, but only 38.57 billion are read for each token. That split is the whole point of the design: it costs the memory of a large model and the speed of a small one. It compresses its cache into a latent vector, which keeps the cache small at long context.

General chatReasoningFrontier scale

Parameters685.4B

Active per token38.57B

Layers61

KV heads128 / 128

Native context160k

Vocabulary129k

About DeepSeek V3.2

DeepSeek V3.2 is a mixture-of-experts model with 685.4B parameters, 38.57B of them active for each token, built from 61 layers with a hidden size of 7168. Its repository was published on 2025-12-01. The license listed on Hugging Face is mit. It compresses its cache into a latent vector, which keeps the cache small at long context.

Memory needed, by quantization

At an 8k context window. Quality is our rough ranking of how much the compression costs you: anything at or above Q5 is hard to tell apart from the original in normal use.

Quantization	File size	KV cache	Total VRAM	Quality	What it costs you
Q8_0	678.22 GB	0.54 GB	678.35 GB	99%	Lossless in practice. Use it when the memory is there.
Q6_K	523.43 GB	0.54 GB	523.77 GB	98%	Very close to Q8 for two thirds of the size.
Q5_K_M	454.01 GB	0.54 GB	454.44 GB	96%	The quality-first choice when Q6 will not fit.
Q4_K_M	385.39 GB	0.54 GB	385.91 GB	93%	The default. Best size-to-quality ratio for local use.
IQ4_XS	339.11 GB	0.54 GB	339.7 GB	91%	Importance-matrix 4-bit. Q4_K_S quality, smaller file.
Q3_K_M	311.98 GB	0.54 GB	312.6 GB	86%	Degradation starts to show. A way to fit one size up.
IQ3_XXS	244.16 GB	0.54 GB	244.87 GB	79%	Aggressive. Only worth it on very large models.

What a longer conversation costs

Same model at Q4_K_M, only the context window changes. This model compresses its cache into a latent vector, which is why the numbers stay small at long context.

Context	KV cache	Total VRAM	Fits in 8 GB	Fits in 12 GB	Fits in 24 GB
4k	0.27 GB	385.65 GB	no	no	no
8k	0.54 GB	385.91 GB	no	no	no
16k	1.07 GB	386.45 GB	no	no	no
32k	2.14 GB	387.52 GB	no	no	no
64k	4.29 GB	389.67 GB	no	no	no
128k	8.58 GB	393.96 GB	no	no	no

The "fits" columns allow for the roughly 0.8 GB Windows keeps for the desktop.

Which hardware runs DeepSeek V3.2

0 of 118 consumer devices run it at a quantization worth using, at an 8k context window. Another 1 can load it only by compressing the weights far enough to damage the model, marked with a warning below.

Device	Memory	Verdict	Quantization	Used	Speed
Apple M3 Ultra 256GB Apple	256 GB	Runs great	IQ2_XXS !	165.38 GB	33 t/s
Apple M2 Ultra 192GB Apple	192 GB	Just fits	n/a !	165.38 GB	32.3 t/s
Apple M3 Ultra 512GB Apple	512 GB	Just fits	n/a	455.02 GB	12.4 t/s
RTX PRO 6000 Blackwell NVIDIA	96 GB	Partial offload 61/61	n/a	385.91 GB	3.1 t/s
GeForce RTX 5090 NVIDIA	32 GB	Partial offload 61/61	n/a	385.91 GB	2.6 t/s
RTX 6000 Ada Generation NVIDIA	48 GB	Partial offload 61/61	n/a	385.91 GB	2.6 t/s
RTX A6000 NVIDIA	48 GB	Partial offload 61/61	n/a	385.91 GB	2.6 t/s
Radeon PRO W7900 AMD	48 GB	Partial offload 61/61	n/a	385.91 GB	2.6 t/s
GeForce RTX 4090 NVIDIA	24 GB	Partial offload 61/61	n/a	385.91 GB	2.5 t/s
GeForce RTX 3090 Ti NVIDIA	24 GB	Partial offload 61/61	n/a	385.91 GB	2.5 t/s
GeForce RTX 3090 NVIDIA	24 GB	Partial offload 61/61	n/a	385.91 GB	2.5 t/s
GeForce RTX 5090 Laptop NVIDIA	24 GB	Partial offload 61/61	n/a	385.91 GB	2.5 t/s
GeForce RTX 5080 NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
GeForce RTX 5070 Ti NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
GeForce RTX 4080 SUPER NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
GeForce RTX 4080 NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
GeForce RTX 4070 Ti SUPER NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
GeForce RTX 3080 Ti NVIDIA	12 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
GeForce RTX 3080 12GB NVIDIA	12 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
GeForce RTX 5080 Laptop NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
RTX A5000 NVIDIA	24 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
Radeon RX 7900 XTX AMD	24 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
Radeon RX 7900 XT AMD	20 GB	Partial offload 61/61	n/a	385.91 GB	2.4 t/s
GeForce RTX 5070 NVIDIA	12 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
GeForce RTX 5060 Ti 16GB NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s

Show 35 more devices

GeForce RTX 4070 Ti NVIDIA	12 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
GeForce RTX 4070 SUPER NVIDIA	12 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
GeForce RTX 4070 NVIDIA	12 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
GeForce RTX 4090 Laptop NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
RTX A4000 NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
Radeon RX 9070 XT AMD	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
Radeon RX 9070 AMD	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
Radeon RX 7900 GRE AMD	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
Radeon RX 7800 XT AMD	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
Radeon RX 6900 XT AMD	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
Radeon RX 6800 AMD	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
Arc A770 16GB Intel	16 GB	Partial offload 61/61	n/a	385.91 GB	2.3 t/s
GeForce RTX 3060 12GB NVIDIA	12 GB	Partial offload 61/61	n/a	385.91 GB	2.2 t/s
GeForce RTX 2060 12GB NVIDIA	12 GB	Partial offload 61/61	n/a	385.91 GB	2.2 t/s
GeForce RTX 4080 Laptop NVIDIA	12 GB	Partial offload 61/61	n/a	385.91 GB	2.2 t/s
Radeon RX 7700 XT AMD	12 GB	Partial offload 61/61	n/a	385.91 GB	2.2 t/s
Radeon RX 6700 XT AMD	12 GB	Partial offload 61/61	n/a	385.91 GB	2.2 t/s
Arc B580 Intel	12 GB	Partial offload 61/61	n/a	385.91 GB	2.2 t/s
GeForce RTX 4060 Ti 16GB NVIDIA	16 GB	Partial offload 61/61	n/a	385.91 GB	2.1 t/s
Radeon RX 7600 XT AMD	16 GB	Partial offload 61/61	n/a	385.91 GB	2.1 t/s
GeForce RTX 5060 Ti 8GB NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 5060 NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 4060 Ti 8GB NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 4060 NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 3080 10GB NVIDIA	10 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 3070 Ti NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 3070 NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 3060 Ti NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 3050 8GB NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 2080 Ti NVIDIA	11 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce GTX 1080 Ti NVIDIA	11 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 4070 Laptop NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
GeForce RTX 4060 Laptop NVIDIA	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
Arc B570 Intel	10 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s
Arc A750 Intel	8 GB	Partial offload 1/61	n/a	385.91 GB	1.4 t/s

How it compares

Other models in the same family or of a similar size, on a 12 GB card and a 24 GB card, each at the best quantization that fits.

Model	Weights at 4 bits	GeForce RTX 3060 12GB	GeForce RTX 4090
DeepSeek V3.2	385.4 GB	does not fit well	does not fit well
DeepSeek R1	384.9 GB	does not fit well	does not fit well
DeepSeek Coder V2 Lite Instruct	8.8 GB	75.7 t/s at Q5_K_M	147.7 t/s at Q8_0
DeepSeek Coder 7B Instruct V1.5	3.9 GB	26 t/s at Q8_0	72.9 t/s at Q8_0
GLM 5.2	423.6 GB	does not fit well	does not fit well
Kimi K2.6	577.4 GB	does not fit well	does not fit well

What to buy to run DeepSeek V3.2

The cheapest hardware that runs it at a quantization worth using and a speed you would not resent, at an 8k context window.

Cheapest that works

Apple M3 Ultra 512GB

$9,499 whole machine

IQ4_XS · uses 340.13 GB of its 512 GB · about 16.5 tokens/s

If it just has to run

RTX PRO 6000 Blackwell

$8,500

Q3_K_M · about 4.1 tokens/s. Cheaper because it runs the model at a lower precision.

Device	Price	Memory	Runs it at	Speed
Apple M3 Ultra 512GB Apple · whole machine	$9,499	512 GB	IQ4_XS	16.5 t/s	Check price

Indicative prices reviewed 2026-09-01; used prices are marketplace typical. Options that cost more than a cheaper one with no more memory and no more speed are hidden. Price links are Amazon affiliate links. Change the standard, the context window or the budget in the buying tool.

These numbers also cover

Fine-tunes share their base model's architecture, so they need exactly the same memory. If you are looking for one of these, the figures above apply unchanged.

deepseek-ai/DeepSeek-V3.2-Exp

Where it sits in the catalog

Source: deepseek-ai/DeepSeek-V3.2 on Hugging Face. Downloaded 1.5M times in the last month. Published 2025-12-01. Architecture figures are read from the repository's own configuration file, so they move when the model does.

DeepSeek R1

684.53B (38.57B active) · 160k context

DeepSeek Coder V2 Lite Instruct

15.71B (2.74B active) · 160k context

DeepSeek Coder 7B Instruct V1.5

6.91B · 4k context

DeepSeek Coder 6.7B Instruct

6.74B · 16k context

Questions people ask about DeepSeek V3.2

How much VRAM does DeepSeek V3.2 need?

At Q4_K_M with an 8k context window, DeepSeek V3.2 needs about 385.91 GB: 384.87 GB of weights, 0.54 GB of KV cache and 0.51 GB of overhead. At the highest-quality file it needs about 678.35 GB.

What is the cheapest GPU that runs DeepSeek V3.2?

No consumer device in our catalog fits this model entirely.

How much does a longer context window add for DeepSeek V3.2?

On the memory side, a 32k window adds 2.1 GB of KV cache on top of the weights, and the cache grows more slowly than a conventional model because of its attention design. The longest window the model supports natively is 160k tokens.