Organization

DeepSeek

DeepSeek is an artificial-intelligence research and product organization founded in Hangzhou in 2023 that develops large language models and related services.[1]independent sourceWhat is DeepSeek, the Chinese AI company upending the stock market?Associated PressThe Associated Press identifies DeepSeek as a Hangzhou-founded AI startup established in 2023; DeepSeek’s official research page and releases document its AI model research and products.CheckedOpen original source [2]first-party sourceDeepSeekDeepSeekDeepSeek’s own site links its web assistant, app, API platform, API documentation and research releases; the wording describes those documented activities without repeating promotional language.CheckedOpen original source

Last reviewedMarkdown

On this page

01

Core facts

Entity type
Organization
Legal name
Hangzhou DeepSeek Artificial Intelligence Co., Ltd.[3]first-party sourceDeepSeek Privacy PolicyDeepSeekDeepSeek’s privacy policy explicitly names Hangzhou DeepSeek Artificial Intelligence Co., Ltd. as the controller providing and controlling the covered services; this is scoped to those services and is not a claim about every affiliate.CheckedOpen original source
Country
China[1]independent sourceWhat is DeepSeek, the Chinese AI company upending the stock market?Associated PressThe Associated Press describes DeepSeek as founded in Hangzhou, China; the company privacy policy says the service controller has its registered address in China.CheckedOpen original source
Official website
api-docs.deepseek.com[4]first-party sourceDeepSeek API DocsDeepSeekThe official DeepSeek API docs are published by DeepSeek and describe programmatic access.CheckedOpen original source

02

What is DeepSeek?[2]first-party sourceDeepSeekDeepSeekDeepSeek’s current home page links the web assistant, app, API platform, API documentation, Harness and its research/news area; the privacy policy identifies the service controller.CheckedOpen original source

DeepSeek develops large language models and provides a web assistant, mobile applications, desktop software and an API. Its research releases include general-purpose language models, code-focused models and reasoning-oriented model families. The company’s current privacy policy identifies Hangzhou DeepSeek Artificial Intelligence Co., Ltd. as the controller for the covered services. Reporting describes DeepSeek’s research origins in a group planned by High-Flyer, while available evidence does not establish that High-Flyer is its current legal parent.

03

History

  1. In 2023 High-Flyer announced plans for a new independent AI research group; Reuters identifies that group as DeepSeek.[5]independent sourceHigh-Flyer, the AI quant fund behind China’s DeepSeekReuters via Investing.comReuters quotes High-Flyer’s 2023 announcement of an independent research group and identifies DeepSeek as the resulting group; the statement does not by itself establish current legal ownership.CheckedOpen original source

  2. DeepSeek was founded in Hangzhou in 2023; the Associated Press reports that its first AI large language model followed later that year.[1]independent sourceWhat is DeepSeek, the Chinese AI company upending the stock market?Associated PressThe Associated Press dates the startup’s founding in Hangzhou to 2023 and says its first AI large language model was released later that year.CheckedOpen original source

  3. DeepSeek released its first AI large language model later in 2023.[1]independent sourceWhat is DeepSeek, the Chinese AI company upending the stock market?Associated PressThe Associated Press reports that the company released its first AI large language model later in the year it was founded.CheckedOpen original source

  4. The DeepSeek-V3 technical report was posted to arXiv on 27 December 2024.[6]structured public data sourceDeepSeek-V3 Technical ReportarXiv / DeepSeek-AIThe arXiv record dates the DeepSeek-V3 report submission to 27 December 2024.CheckedOpen original source

  5. The DeepSeek-Coder technical paper was posted to arXiv on 25 January 2024.[7]structured public data sourceDeepSeek-Coder: When the Large Language Model Meets ProgrammingarXiv / DeepSeek-AIThe arXiv record dates the DeepSeek-Coder paper submission to 25 January 2024.CheckedOpen original source

  6. The DeepSeek-Coder technical paper was posted to arXiv on 25 January 2024.[7]structured public data sourceDeepSeek-Coder: When the Large Language Model Meets ProgrammingarXiv / DeepSeek-AIThe arXiv record dates the DeepSeek-Coder paper submission to 25 January 2024.CheckedOpen original source

  7. The DeepSeek-V3 technical report was posted to arXiv on 27 December 2024.[6]structured public data sourceDeepSeek-V3 Technical ReportarXiv / DeepSeek-AIThe arXiv record dates the DeepSeek-V3 report submission to 27 December 2024.CheckedOpen original source

  8. DeepSeek announced V3.2 and V3.2-Speciale on 1 December 2025.[8]first-party sourceDeepSeek-V3.2 releaseDeepSeekThe official release page dates the two model announcements to 1 December 2025.CheckedOpen original source

  9. DeepSeek released R1 and R1-Zero and published their technical report in January 2025.[9]first-party sourceDeepSeek-R1 ReleaseDeepSeekDeepSeek’s official release is dated 20 January 2025 and links the technical report; the arXiv record was submitted on 22 January 2025.CheckedOpen original source

  10. DeepSeek released R1 and R1-Zero and published their technical report in January 2025.[9]first-party sourceDeepSeek-R1 ReleaseDeepSeekDeepSeek’s official release is dated 20 January 2025 and links the technical report; the arXiv record was submitted on 22 January 2025.CheckedOpen original source

  11. DeepSeek announced V3.2 and V3.2-Speciale on 1 December 2025.[8]first-party sourceDeepSeek-V3.2 releaseDeepSeekThe official release page dates the two model announcements to 1 December 2025.CheckedOpen original source

  12. DeepSeek announced V4.1-Flash on 10 September 2026.[10]first-party sourceIntroducing DeepSeek-V4.1-FlashDeepSeekThe official release page dates the V4.1-Flash announcement to 10 September 2026.CheckedOpen original source

  13. DeepSeek announced the V4 Preview on 24 April 2026.[11]first-party sourceDeepSeek-V4 PreviewDeepSeekThe official V4 release page dates the preview announcement to 24 April 2026.CheckedOpen original source

  14. DeepSeek announced the V4 Preview on 24 April 2026.[11]first-party sourceDeepSeek-V4 PreviewDeepSeekThe official V4 release page dates the preview announcement to 24 April 2026.CheckedOpen original source

  15. DeepSeek announced V4.1-Flash on 10 September 2026.[10]first-party sourceIntroducing DeepSeek-V4.1-FlashDeepSeekThe official release page dates the V4.1-Flash announcement to 10 September 2026.CheckedOpen original source

04

Models and model families

Deepseek Coder
The DeepSeek-Coder research series comprises code-focused models from 1.3B to 33B parameters; its paper describes code generation and infilling research.[7]structured public data sourceDeepSeek-Coder: When the Large Language Model Meets ProgrammingarXiv / DeepSeek-AIThe DeepSeek-Coder paper identifies the 1.3B–33B model range and explains its programming-language-model training and code completion tasks.CheckedOpen original source
Deepseek LLM
The DeepSeek-LLM research series includes 7B and 67B parameter language-model configurations and chat variants, described in a January 2024 technical paper.[12]structured public data sourceDeepSeek LLM: Scaling Open-Source Language Models with LongtermismarXiv / DeepSeek-AIThe DeepSeek-LLM technical paper documents the 7B and 67B model scales and chat variants; this is a research-paper description, not a current service availability claim.CheckedOpen original source
Deepseek V3
DeepSeek-V3 is a large language model documented in a December 2024 technical report; the report describes a mixture-of-experts architecture and multi-token prediction training objective.[6]structured public data sourceDeepSeek-V3 Technical ReportarXiv / DeepSeek-AIThe DeepSeek-AI technical report describes V3’s architecture and training approach. Technical specifications are attributed to the research team’s paper, not presented as independent performance validation.CheckedOpen original source
Reasoning
The DeepSeek-R1 research release includes R1-Zero, R1 and six distilled model sizes; its January 2025 paper describes reinforcement-learning and multi-stage training approaches.[13]structured public data sourceDeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement LearningarXiv / DeepSeek-AIThe R1 technical paper documents R1-Zero, R1 and six distilled sizes and explains their training approaches; the associated release records publication on 20 January 2025.CheckedOpen original source
V3 2
DeepSeek-V3.2 and V3.2-Speciale were announced on 1 December 2025; the company described V3.2 as available through its app, web and API and Speciale as API-only at launch.[8]first-party sourceDeepSeek-V3.2 releaseDeepSeekDeepSeek’s dated release names both variants and distinguishes their launch channels. The page’s comparative benchmark statements are not repeated here.CheckedOpen original source
V4
DeepSeek-V4 Preview was released on 24 April 2026 with model weights and an API update; the company stated a one-million-token context window for its official services at launch.[11]first-party sourceDeepSeek-V4 PreviewDeepSeekDeepSeek’s dated V4 announcement records the preview release, open-weight distribution, API availability and the company-stated context length at launch.CheckedOpen original source
V4 1 Flash
DeepSeek-V4.1-Flash was announced on 10 September 2026; the company says it has native visual understanding and is available through the DeepSeek API as deepseek-flash.[10]first-party sourceIntroducing DeepSeek-V4.1-FlashDeepSeekThe current official release dates V4.1-Flash to 10 September 2026, describes native visual understanding and gives the API model identifier. These are company-published specifications.CheckedOpen original source

05

Contributions and work

Focus
Research and releases cover large language models, code-focused language models, reasoning models and agent-oriented capabilities.[14]first-party sourceResearch & NewsDeepSeekDeepSeek’s official research/news index links separate releases and technical reports for V4.1, V4, V3.2, V3.1, R1 and V3; the Coder and early LLM series are documented in their research papers.CheckedOpen original source

06

Products

API
An API whose format is compatible with OpenAI and Anthropic.[4]first-party sourceDeepSeek API DocsDeepSeekThe official DeepSeek API docs say the API format is compatible with OpenAI and Anthropic.CheckedOpen original source
Desktop Harness
DeepSeek Harness is a desktop application; its download page lists macOS (Apple silicon, macOS 13 or later) and Windows downloads.[15]first-party sourceDownload DeepSeekDeepSeekThe official download page identifies DeepSeek Harness for desktop and provides platform-specific macOS and Windows downloads.CheckedOpen original source
Web Assistant
DeepSeek’s official website links a web-based assistant at chat.deepseek.com.[2]first-party sourceDeepSeekDeepSeekThe official home page exposes DeepSeek Web as a separate service link to chat.deepseek.com.CheckedOpen original source

07

Services

API
DeepSeek offers a developer API with model-specific documentation and update notices; as of September 2026, its official V4.1-Flash release names deepseek-flash as the API model identifier.[10]first-party sourceIntroducing DeepSeek-V4.1-FlashDeepSeekThe official model release says V4.1-Flash is live on the API and specifies deepseek-flash; the API documentation provides model and protocol updates.CheckedOpen original source

08

Operating model

Research Origin
DeepSeek emerged in 2023 as an independent AI research group planned by High-Flyer; reporting identifies Liang Wenfeng as its founder and leader.[5]independent sourceHigh-Flyer, the AI quant fund behind China’s DeepSeekReuters via Investing.comReuters reports that High-Flyer announced a new independent AI research group in 2023, identifies that group as DeepSeek, and says Liang Wenfeng founded High-Flyer and led DeepSeek. This describes research origins, not a verified current legal ownership chain.CheckedOpen original source
Services
DeepSeek provides consumer web and mobile assistant services and a separate developer API platform; the API platform publishes model documentation and service updates.[2]first-party sourceDeepSeekDeepSeekThe official home page links the Web, App and API Platform as distinct service channels, and the linked API documentation maintains a model/update reference.CheckedOpen original source

10

Sources

Reference numbers stay stable across this page and each source appears once, ordered by first visible use.

  1. DeepSeek

    DeepSeek · first-party source

  2. DeepSeek Privacy Policy

    DeepSeek · first-party source

  3. DeepSeek API Docs

    DeepSeek · first-party source

  4. High-Flyer, the AI quant fund behind China’s DeepSeek

    Reuters via Investing.com · independent source

  5. DeepSeek-V3 Technical Report

    arXiv / DeepSeek-AI · structured public data source

  6. DeepSeek-Coder: When the Large Language Model Meets Programming

    arXiv / DeepSeek-AI · structured public data source

  7. DeepSeek-V3.2 release

    DeepSeek · first-party source

  8. DeepSeek-R1 Release

    DeepSeek · first-party source

  9. Introducing DeepSeek-V4.1-Flash

    DeepSeek · first-party source

  10. DeepSeek-V4 Preview

    DeepSeek · first-party source

  11. DeepSeek LLM: Scaling Open-Source Language Models with Longtermism

    arXiv / DeepSeek-AI · structured public data source

  12. Research & News

    DeepSeek · first-party source

  13. Download DeepSeek

    DeepSeek · first-party source

  14. DeepSeek — Wikidata

    Wikidata · structured public data source

  15. Introducing DeepSeek App

    DeepSeek · first-party source

  16. DeepSeek - AI Assistant

    Google Play listing (developer: DeepSeek) · independent source

Found something that needs review?

Suggest a correction