MODEL LINEAGE
中文
Model index
DeepSeek·DeepSeek V / R

DeepSeek-V3.2

A general model unifying non-thinking and thinking modes with sparse attention and everyday agent capability.

LegacyLanguageReasoningCodingAgentOpen weights
MODEL NUMBER100
First public date
2025-12-01
Organization
DeepSeek
Family
DeepSeek V / R
01

Specifications

Only published specifications are shown.

Total parameters
671B
Active parameters
37B
Architecture
MoE
Context
128,000 tokens
Input modalities
Text
Open weights
Yes
License
MIT
Status
Legacy
02

Overview

Release dates

— DeepSeek released open weights for V3.2 and its high-reasoning Speciale model. ↗

03

Specifications alongside predecessor

Compared with DeepSeek-V3.1-Terminus ↗

PreviousDeepSeek-V3.1-Terminus— · — ctx
CurrentDeepSeek-V3.2671B · 128K ctx
04

Architecture & capabilities

Attention

DSA

Model capabilities

Hybrid reasoningAgentCodingTool use
05

Release history

Open weights

DeepSeek released open weights for V3.2 and its high-reasoning Speciale model.

View source
06

Related models

Select a card to view the related model.

PredecessorDeepSeek-V3.1-TerminusView details →SuccessorDeepSeek-V4-Pro-PreviewView details →VariantDeepSeek-V3.2-SpecialeView details →
07

Sources

01
DeepSeek API change log: DeepSeek-V3.2DeepSeek · Published 2025-12-01
Official · Verified
Suggest a correction

Copy this template, add the field, proposed value, and primary source, and share it with the archive maintainer.