General · EdgepediaSeptember 17, 20265 min read

DeepSeek (chatbot)

DeepSeek is a free AI chatbot made by the Chinese company of the same name, offered as a website at chat.deepseek.com and as mobile apps for iOS and Android, and powered by the company's own large language models¹^•². Its mobile app launched on January 15, 2025, running DeepSeek-V3, and in late January 2025, the month the company released its R1 reasoning model, it rose to No. 1 on Apple's App Store and Google Play¹^•³. TechCrunch partly attributes an 18% fall in Nvidia's stock that month to DeepSeek's success³. The service has drawn government restrictions over data security and censors answers on topics such as Tiananmen Square and Taiwan³.

Key fact	Detail
Developer	DeepSeek, an AI lab started in 2023 by the hedge fund High-Flyer, which remains an investor³
Data controller	Hangzhou DeepSeek Artificial Intelligence Co., Ltd.⁴
App launch	January 15, 2025, on the App Store, Google Play, and other Android markets¹
Price	Free, with no ads or in-app purchases¹
Current models	DeepSeek-V4-Pro in Expert Mode (from August 13, 2026); DeepSeek-V4.1-Flash-1017 (from October 17, 2026)⁵^•⁶
Context window	1 million tokens, the default across DeepSeek's official services since April 2026²
Data storage	Personal data collected, processed, and stored in the People's Republic of China⁴

Origins

DeepSeek grew out of High-Flyer Capital Management, a Chinese quantitative hedge fund co-founded in 2015 by Liang Wenfeng that began using AI for trading; High-Flyer set up DeepSeek as a research lab in 2023 and later spun it off as a separate company³. In November 2023 it released DeepSeek Coder, DeepSeek LLM, and DeepSeek Chat³. The company has trained on its own data center clusters, and under US export controls it used Nvidia H800 chips, a weaker version of the H100, to train a recent model³.

Features

The January 2025 app let users sign in with email, Google, or an Apple ID, synced chat history across devices, and offered web search, a Deep-Think reasoning mode, and file uploads with text extraction¹. Since the V4 preview of April 24, 2026, the web and app versions have offered an Expert Mode and an Instant Mode, and the underlying models support both thinking and non-thinking modes over a context of 1 million tokens². DeepSeek's homepage lists deep thinking and intelligent search as the chatbot's main features and also offers a desktop client, DeepSeek Harness⁷.

Models behind the chatbot

Successive models have replaced one another quickly. DeepSeek-V3 arrived in December 2024 and the R1 reasoning model in January 2025, which DeepSeek said matched OpenAI's o1 on key benchmarks; an updated R1 followed in May 2025 and an experimental V3.2 in September 2025 aimed at cheaper long-context use³. The V4 preview introduced two open-weight models: V4-Pro, with 1.6 trillion total parameters of which 49 billion are active for each token, and V4-Flash, with 284 billion total and 13 billion active². V4-Pro became generally available in the app and on the web through Expert Mode on August 13, 2026, with adjustable reasoning effort at low, high, and max settings⁵. On October 17, 2026 DeepSeek put V4.1-Flash-1017 on its website, app, and API, describing it as faster, higher scoring on benchmarks, and terser than the version before it⁶.

DeepSeek prices its products well below market rates and gives some away, and it has declined outside investment; some experts dispute its claims about how cheaply it trains its models³. Its models carry permissive licenses that allow commercial use, and developers had built more than 500 derivative models with 2.5 million combined downloads, according to Hugging Face's chief executive Clem Delangue³.

Usage

After its January 2025 surge the chatbot's audience fell back: Similarweb measured its daily visits in March 2025 at 25% below February, in a month when ChatGPT passed 500 million weekly active users³.

Censorship and data concerns

China's internet regulator tests the models to ensure their responses "embody core socialist values," and in the chatbot R1 declines questions about Tiananmen Square or Taiwan's autonomy³. DeepSeek's privacy policy, last updated February 10, 2026, says it collects users' typed or spoken inputs, uploaded files and images, chat history, device and network details, and approximate location from IP addresses, and stores personal data in China⁴.

Restrictions followed in several countries. Italy's data protection authority, the Garante, ordered DeepSeek to block the chatbot there in January 2025 after it failed to answer concerns about its privacy policy⁸. In February 2025 South Korea's Personal Information Protection Commission suspended new downloads from local app stores, leaving the web service reachable, after DeepSeek acknowledged it had partly neglected the country's data protection law⁸. The Pentagon moved to block the service in January 2025 after employees connected to Chinese servers, New York State banned it on government devices, and in May 2025 Microsoft's president Brad Smith told a Senate hearing that Microsoft employees may not use the app, citing data security and propaganda³. A Chinese foreign ministry spokesperson said in February 2025 that Beijing would never ask companies to collect or store data in breach of laws⁸.

References

Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Modern AI: foundation models, generative AI, and the AI industry › AI companies, people, and products › AI products and assistants › Chat assistants and AI search

Initially written Sep 17, 2026 · Reviewed: Sep 30, 2026; Oct 25, 2026 · Edited: Sep 30, 2026; Oct 25, 2026 · Last review: Oct 25, 2026

Notice something wrong?