DeepSeek Funding History

DeepSeek develops language models, releases research and model weights, and operates a chat assistant and API. The Hangzhou company grew out of founder Liang Wenfeng’s work at quantitative investment firm High-Flyer.

Company Overview

DeepSeek logo

DeepSeek

Founded 2023 · Hangzhou, China

Official Website

DeepSeek Founding Team

Founder Liang Wenfeng brought a quantitative-research and computing background from High-Flyer. DeepSeek publishes technical reports and model implementations alongside its hosted services.

DeepSeek Products

  • DeepSeek

    Use DeepSeek’s hosted conversational assistant.

  • DeepSeek API

    Integrate DeepSeek’s hosted models into software through its developer API.

  • DeepSeek

    Access published model weights and technical research for development and deployment.

DeepSeek Company Development

A Research Company Grows Out of High-Flyer

Liang Wenfeng founded DeepSeek in 2023 after building High-Flyer’s quantitative investment business. Contemporary reporting and a July 2025 Congressional Research Service overview describe the hedge fund as its financial backer. That history differs from a conventional sequence of disclosed venture rounds; no external financing amount is inferred from computing resources or the founder’s assets.

V3 Links Open Weights with Efficiency Research

The V3 release combined public model weights and a technical report describing the model’s training approach. Its reported compute expenditure became a prominent part of the company’s public profile. That figure describes a specific training calculation, rather than the company’s total operating expenditure or equity financing.

R1 Makes Reasoning a Separate Release Strategy

DeepSeek-R1 expanded the company’s profile in reasoning research. The original release includes both DeepSeek’s own reasoning models and smaller models distilled into Qwen and Llama architectures. That distinction matters commercially: releasing a compact derivative does not mean DeepSeek pretrained its underlying foundation model from scratch.

Distribution Spans Weights, Chat and the API

DeepSeek distributes research checkpoints separately from its hosted products. Its current API documentation lists Flash and V4 Pro access and explains that some older endpoint names now route to a newer model. V3 and R1 remain important earlier releases, while the hosted API has moved to newer models.

FAQ

Who founded DeepSeek?

Liang Wenfeng founded DeepSeek after building quantitative investment firm High-Flyer.

Does DeepSeek only operate a chatbot?

No. It develops and releases models and also provides a consumer assistant and developer API.

Are DeepSeek’s models and its API the same product?

A downloadable model is a research or deployment artifact. The hosted API is a service whose model names and serving behavior can change.

What does DeepSeek’s V3 training-cost figure measure?

It is an estimate of compute expenditure for a specific model-training calculation, rather than DeepSeek’s total operating costs or money raised from investors.

How has DeepSeek financed its research?

DeepSeek’s early research was backed by High-Flyer. The cited sources do not establish a completed, separately disclosed external funding round or a verified amount raised from outside investors.