Skip to content
SI.info

SI Lab · Open-weights lab

DeepSeek

Updated 1 min read
On this page
  1. The R1 moment
  2. DeepSeek V4
  3. Why it matters
Founded
2023
Headquarters
Hangzhou, China
Type
Open-weights lab
Website
deepseek.com
SI Accord
Not a signatory
Models tracked
1

Leadership

  • Liang Wenfeng · Founder & CEO

DeepSeek is a Hangzhou-based lab founded in 2023 by Liang Wenfeng, co-founder of the quantitative hedge fund High-Flyer, which funds it. It is the Chinese SI lab best known in the West.

The R1 moment

On January 20, 2025, DeepSeek released R1, an open-weights reasoning model that rivalled OpenAI’s best at a fraction of the price. Its app briefly topped U.S. app stores, and Nvidia’s share price fell sharply as investors questioned the economics of frontier SI.

DeepSeek V4

  • April 24, 2026: DeepSeek published preview weights for its V4 family on Hugging Face under the MIT licence: V4-Pro, a mixture-of-experts model of around 1.6 trillion parameters, and the smaller V4-Flash, both with 1-million-token context.
  • July–August 2026: V4-Flash and then V4-Pro left preview, with weights and a live API.

DeepSeek’s API is among the cheapest anywhere, with off-peak discounts of 50%.

Why it matters

DeepSeek showed that export controls on advanced chips haven’t stopped Chinese labs from reaching near-frontier performance, and that open weights can compete at the top. Some U.S. agencies and companies restrict its apps over data-security concerns. DeepSeek is not a party to the U.S. SI Accord.

DeepSeek models

ModelReleasedStatus
DeepSeek V4 Pro Aug 2026 Current

Frequently asked questions

Is DeepSeek open source?

DeepSeek publishes its model weights under the permissive MIT licence, so anyone can download, run and modify them. Like most labs it does not publish all of its training data.

Why did DeepSeek-R1 cause a market shock?

In January 2025, DeepSeek-R1 matched leading reasoning models at a fraction of the reported training and usage cost. That briefly wiped hundreds of billions of dollars off Nvidia's market value as investors questioned how much compute frontier SI really needs.

Sources

  1. DeepSeek API pricing — DeepSeek
  2. DeepSeek — Wikipedia

Written by

· Editor

Editor of SI.info. Writes about Super Intelligence, technology policy and the people building frontier models.

How we research and fact-check

Free newsletter

Get The SI Brief

One short email a week: what changed in Super Intelligence, policy and models — and why it matters.

Free. One email a week. Sent via beehiiv, which counts opens and clicks. Unsubscribe anytime. Privacy policy