What are the hardware requirements for DeepSeek V3.2?

DeepSeek V3.2 requires approximately 172 GB of VRAM and supports a context window of 160K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed DeepSeek V3.2 and how large is it?

DeepSeek V3.2 is developed by unsloth, a model with 685.4b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does DeepSeek V3.2 perform on standard benchmarks?

DeepSeek V3.2 has the following published scores: LMSYS ELO 1425. Compare against reference models on this page or on the LLM Explorer leaderboards.

DeepSeek V3.2 by unsloth — VRAM 172GB, 160K context

Name: DeepSeek V3.2
Author: unsloth

DeepSeek V3.2 is an open-source language model by unsloth. Features: 685.4b LLM, VRAM: 172GB, Context: 160K, License: mit, LLM Explorer Score: 0.36, ELO: 1425.

Base model:deepseek-ai/deepsee... Base model:finetune:deepseek-a... Conversational Deepseek v32 Endpoints compatible Fp8 Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/unsloth/DeepSeek-V3.2

DeepSeek V3.2 Benchmarks

LMSys ELO 1425 vs 1272 (so35) 12%

^nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

What is the LLM Explorer Rank (Score)

DeepSeek V3.2 Parameters and Internals

LLM Name	DeepSeek V3.2
Repository 🤗	https://huggingface.co/unsloth/DeepSeek-V3.2
Base Model(s)	deepseek-ai/DeepSeek-V3.2-Exp-Base deepseek-ai/DeepSeek-V3.2-Exp-Base
Model Size	685.4b
Required VRAM	172 GB
Updated	2026-07-30
Maintainer	unsloth
Model Type	deepseek_v32
Model Files	5.2 GB: 1-of-163 4.3 GB: 2-of-163 4.3 GB: 3-of-163 4.3 GB: 4-of-163 4.3 GB: 5-of-163 4.3 GB: 6-of-163 4.3 GB: 7-of-163 4.3 GB: 8-of-163 4.3 GB: 9-of-163 4.3 GB: 10-of-163 4.3 GB: 11-of-163 1.5 GB: 12-of-163 4.3 GB: 13-of-163 4.3 GB: 14-of-163 4.3 GB: 15-of-163 4.3 GB: 16-of-163 4.3 GB: 17-of-163 4.3 GB: 18-of-163 4.3 GB: 19-of-163 4.3 GB: 20-of-163 4.3 GB: 21-of-163 4.3 GB: 22-of-163 4.3 GB: 23-of-163 4.3 GB: 24-of-163 4.3 GB: 25-of-163 4.3 GB: 26-of-163 4.3 GB: 27-of-163 4.3 GB: 28-of-163 4.3 GB: 29-of-163 4.3 GB: 30-of-163 4.3 GB: 31-of-163 4.3 GB: 32-of-163 4.3 GB: 33-of-163 1.9 GB: 34-of-163 4.3 GB: 35-of-163 4.3 GB: 36-of-163 4.3 GB: 37-of-163 4.3 GB: 38-of-163 4.3 GB: 39-of-163 4.3 GB: 40-of-163 4.3 GB: 41-of-163
Model Architecture	DeepseekV32ForCausalLM
License	mit
Context Length	163840
Model Max Length	163840
Transformers Version	4.44.2
Vocabulary Size	129280
Torch Data Type	bfloat16

Rank the DeepSeek V3.2 capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to DeepSeek V3.2

Best Alternatives	Context / RAM	Downloads	Likes
DeepSeek V3.2 Exp	160K / 180.6 GB	214096	994
DeepSeek V3.2	160K / 176.3 GB	1246445	1463
DeepSeek V3.2 Exp Base	160K / 184.9 GB	426	68
DeepSeek V3.2 Speciale	160K / 176.3 GB	5128	718
DeepSeek Math V2	160K / 184.9 GB	376	705

Note: green Score (e.g. "73.2") means that the model is better than unsloth/DeepSeek-V3.2.