Renaissance Labs

Atlas / Asia

Independent company · Hangzhou, China

DeepSeek

A Hangzhou AI lab founded by Liang Wenfeng and owned by the hedge fund High-Flyer, known for open-weight V3, R1 and V4 models released under the MIT License.

language modelsreasoningChineseEnglish

Hangzhou, China

Founded
2023
Type
Independent company
Based in
Hangzhou, China
Key papers
4
Featured projects
8

About

DeepSeek was founded on 17 July 2023 as a spin-off from the AI research lab of High-Flyer, a Chinese hedge fund that owns and backs it. Liang Wenfeng founded the company and serves as CEO, and it had about 160 employees in 2025.

The lab released DeepSeek-V2 in May 2024, which introduced multi-head latent attention within a mixture-of-experts design, and DeepSeek-V3 in December 2024. DeepSeek-R1 followed on 20 January 2025 alongside a chatbot, with weights under the MIT License. Later releases include V3.1 in August 2025, V3.2 in December 2025 and a V4 preview in April 2026.

Wikipedia reports that DeepSeek closed a $7 billion Series A at a $52 billion post-money valuation in May 2026, ending its reliance on its parent alone. The company publishes weights and technical reports but not training data, and several governments have restricted its use on official systems.

Mission and ambitions

DeepSeek pursues capable general models and publishes weights and technical reports for them, funded originally by High-Flyer and, from 2026, by outside investors as well.

Known for

  1. 01

    DeepSeek-R1 reasoning model

  2. 02

    MIT-licensed open weights

  3. 03

    Multi-head latent attention

  4. 04

    DeepSeek-V3 mixture-of-experts model

Milestones

  1. 2023

    DeepSeek founded

    The company was established as a spin-off of High-Flyer's AI research lab.

  2. 2024

    DeepSeek-V2

    The model introduced multi-head latent attention with a mixture-of-experts architecture.

  3. 2024

    DeepSeek-V3

    A larger mixture-of-experts model was released in December 2024.

  4. 2025

    DeepSeek-R1

    The reasoning model launched on 20 January 2025 and was released under the MIT License.

  5. 2025

    DeepSeek-V3.2

    Released in December 2025, following V3.1 in August 2025.

  6. 2026

    DeepSeek-V4 preview

    Previewed in April 2026, with Flash and Pro versions released later that year.

Open source

GitHubgithub.com/deepseek-ai
Hugging Facehuggingface.co/deepseek-ai

Research papers

Notable releases

Openness

Model weights are published on Hugging Face; licences vary by release (check each model card).

People

  • Liang WenfengFounder and CEO

    Also co-founder of the parent fund High-Flyer.

Listed from the lab's own pages and public profiles.

Contact and community

The official site prints a service email address; code and model questions go through GitHub issues and Hugging Face discussions.

Sources

Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.

Labs nearby