AI Hub
All models
LLMOpen weights

DeepSeek-V2

DeepSeek

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

Superseded by DeepSeek-V3

Specifications

Type
LLM
Access
Open weights
Released
May 6, 2024
Parameters
236B total / 21B active (MoE)
Context window
128,000 tokens
Input
Text
Output
Text

Our take

Introduced multi-head latent attention for cheap inference and undercut the market on price, foreshadowing the V3 and R1 breakthroughs.

Related research

Compare DeepSeek-V2 with

See all DeepSeek-V2 alternatives →