Arrakis / Models / DeepSeek-V3.2

DeepSeek-V3.2 GPU requirements

DeepSeek-V3.2 has 685B parameters. To serve it, plan for about 1,645 GB of GPU memory in 16-bit, 822 GB in 8-bit or 411 GB in 4-bit, including about 20% for the KV cache and activations.

It is a mixture of experts: 256 experts, 8 of them active for each token. All of them stay in GPU memory, so plan memory for all 685B parameters.

Deploy this model Check another model

What it takes to serve it

GPU memory to run it for inference, with the smallest GPU setup that holds it and the lowest price on Arrakis right now.

Precision	Memory and the smallest setup
16-bit	1,645 GB: 8 × B300, from $36.80 per hour for all 8
8-bit	822 GB: 4 × B300, from $18.40 per hour for all 4
4-bit	411 GB: 2 × B300, from $9.20 per hour for all 2

What it takes to fine-tune it

QLoRA trains adapters on a 4-bit frozen model, LoRA on a 16-bit one, and full fine-tuning updates every weight. Figures are before activations, which grow with batch size and sequence length.

Method	Memory and the smallest setup
QLoRA	480 GB: 2 × B300, from $9.20 per hour for all 2
LoRA	1,645 GB: 8 × B300, from $36.80 per hour for all 8
Full fine-tuning	10,966 GB: more than one 8-GPU machine

About the model

From its card and files on Hugging Face. Memory figures here follow the GPU memory guide.

Parameters: 685B
Experts: 256
Experts for each token: 8
Context window (tokens): 163,840
Published in: FP8
License: mit
Released: 2025-12-01
Weights: deepseek-ai/DeepSeek-V3.2

Deploy it on Arrakis

Open Models in the console, choose vLLM, SGLang or llama.cpp, and rent a GPU it fits on. You get an OpenAI-compatible endpoint with its own API key, and root SSH to the machine. Or run it yourself on any machine with enough GPU memory:

vLLM

vllm serve deepseek-ai/DeepSeek-V3.2 --tensor-parallel-size 8

Deploy this model How deploying works

Questions

Live prices for every GPU are on the GPU prices page.

How much GPU memory does DeepSeek-V3.2 need?

To serve it, about 1,645 GB in 16-bit, 822 GB in 8-bit, 411 GB in 4-bit, including about 20% for the KV cache and activations. To fine-tune it, about 480 GB with QLoRA, 1,645 GB with LoRA and 10,966 GB for full fine-tuning with Adam, before activations.

Can I run DeepSeek-V3.2 on one GPU?

No. Even in 4-bit it needs about 411 GB: 2 B300 GPUs.

What does it cost to run DeepSeek-V3.2?

In 16-bit it needs 8 B300 GPUs; in 4-bit, 2 B300 GPUs. Each setup on this page shows the lowest price per GPU-hour on Arrakis right now, and you pay by the hour from a prepaid balance.

How do I deploy DeepSeek-V3.2 on Arrakis?

Open Models in the Arrakis console, paste deepseek-ai/DeepSeek-V3.2, choose vLLM, SGLang or llama.cpp, and rent a GPU it fits on. You get an OpenAI-compatible endpoint with its own API key, and root SSH to the machine.

More models

Every model, and a check for any other.