DeepSeek-V2

Name: DeepSeek-V2
Author: DeepSeek

DeepSeekDeepSeekStatus: legacy

DeepSeek-V2 is a general purpose, foundation, mixture-of-experts AI model developed by DeepSeek. It was released on 2024-05-06. Its tracked lifecycle status is legacy. It supports a context window of 131,072 tokens. It has 236B parameters.

Specifications

Organization: DeepSeek
Model family: DeepSeek
Generation: V2
Status: legacy
Released: 2024-05-06
Model type: General Purpose, Foundation, Mixture-of-Experts
Input modalities: text
Output modalities: text
Context window: 131,072 tokens
Parameters: 236B
License: MIT
Weights: open weights
Official API: Yes
Self-hostable: Yes
Last verified: 2026-08-01

Capabilities

Tool use

Official sources

DeepSeek API Docs Official site Hugging Face

Related models in the DeepSeek family

DeepSeek-V4.1-Flash 2026-09-10 · active
DeepSeek-V4-Flash-Vision-Exp 2026-08-31 · active
DeepSeek-V4-Pro-0813 2026-08-13 · active
DeepSeek-V4-Flash-0731 2026-07-31 · active
DeepSeek-R1 2025-01-20 · active
DeepSeek-R1 2025-01-20 · active
DeepSeek-R1-Distill-Llama-8B 2025-01-20 · active
DeepSeek-R1-Distill-Qwen-32B 2025-01-20 · active
DeepSeek-V3 2024-12-26 · active
DeepSeek-Coder-V2 2024-06-17 · active
DeepSeek-V2 Lite 2024-05-06 · active
DeepSeek-Coder 2024-01-25 · legacy

Interactive filtering and search on this page need JavaScript; every fact above is part of the page itself.