MMOOZZGG.com

Glossary

Who is Liang Wenfeng?

In short

Liang Wenfeng is the founder of DeepSeek, a Chinese company that releases language models with open weights.

People · Liang Wenfeng

MMOOZZGG.com editorsChecked: 1 min read

For a child

Imagine a student who was great at math problems and then gathered friends into a club. Together they built a smart helper. The helper solves a problem step by step and checks itself, and the club shares its recipe with anyone who wants it. Liang Wenfeng gathered exactly such a team, only of adults, and people in many countries use his helper.

In plain words

Liang started in finance: in 2015 he co-founded High-Flyer, a quantitative hedge fund that trades with mathematical models. In 2023 he founded DeepSeek. The company builds large language models and publishes their weights openly, so you can download a model and run it on your own computer. The company described its approach to reasoning with the R1 model: before it answers, it solves the task step by step and checks itself. For example, a student asks such a model to solve a physics problem and sees not only the answer but also the solution steps.

For business

Liang’s work matters to companies because of open weights. The weights of the V4 series are published under the MIT license, which allows commercial use and changes. So you can put the model on a server in Russia and not send data to China or anywhere else. Costs move from paying per request to hardware and a team: large versions need several powerful GPUs, while compact ones cost less. Before launch, test the quality on your own tasks and in Russian, the hardware needs and the license terms of the exact version. If you use the developer’s website or API, check separately where the data is stored. How to choose between your own server and the cloud is covered in the section on sovereign AI.

For a specialist

According to Fortune (27 January 2025), Liang co-founded the quantitative hedge fund High-Flyer in 2015 and founded DeepSeek in 2023. The company’s key paper is “DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning” by DeepSeek-AI; Liang himself submitted it to arXiv on 22 January 2025. The paper later appeared in Nature (vol. 645, pp. 633–638, 2025). It shows that reasoning skill can be developed with reinforcement learning without human-labeled reasoning examples. The model gains self-verification and self-reflection, and the reasoning patterns it finds help strengthen smaller models through distillation. The open-weights policy continued in the V4 series: the DeepSeek-V4-Pro repository on Hugging Face is released with its weights under the MIT license.

Example

An equipment maker wants its engineers to ask a model questions about technical documents without letting the files leave its network. The team installs the open DeepSeek model on its own server and tests how it reasons over questions about diagrams and tolerances.

Related terms

Questions and answers

Who founded DeepSeek? Liang Wenfeng founded DeepSeek in 2023. Before that, in 2015, he co-founded High-Flyer, a quantitative hedge fund.
What is Liang Wenfeng known for in AI? His company releases open-weights models and described the R1 reasoning model in a paper that appeared in Nature in 2025. The weights of the V4 series are released under the MIT license.

Sources

Not sure whether your company needs this?

AI solutions map
Contact us