AI Hub
All models
LLMOpen weights

DeepSeek-V3.2

DeepSeek

Updated May 21, 2026 · Benchmarks via Artificial Analysis, specs & pricing via OpenRouter · Methodology · Report an error

Superseded by DeepSeek-V4-Pro

Specifications

Type
LLM
Access
Open weights
Released
December 1, 2025
License
MIT
Parameters
671B (37B active)
Context window
131,072 tokens
Input
Text
Output
Text
API pricing
$0.25 in · $0.38 out / 1M tokens
Capabilities
Function callingStructured output

Benchmarks

Our take

Introduced DeepSeek Sparse Attention (DSA) for near-linear long-context cost; a high-compute “Speciale” variant reached IMO/IOI gold-medal level.

Related research

Links

Compare DeepSeek-V3.2 with

See all DeepSeek-V3.2 alternatives →