Glossary › Players and products

DeepSeek

A Chinese artificial intelligence company and its namesake chatbot, which in early 2025 disrupted the global AI market with freely downloadable language models developed at a much lower cost than expected.

In plain terms

DeepSeek is both a Hangzhou-based Chinese company and the name of the chatbot it created. You use it the same way as ChatGPT: you type your question, and it answers, writes text, translates, summarizes, or creates program code.

It became famous for showing that a near-state-of-the-art model does not necessarily require as much money and as many top-tier chips as the American giants spend. It is like an obscure car manufacturer launching a car comparable to those of major brands for a fraction of the price, where anyone can freely modify the finished car. The much-cited development cost of approximately 6 million dollars, however, covers only the final training run; the total expenditure was much higher.

The other difference is openness. DeepSeek makes the 'weights' of its models, meaning the trained model itself, freely downloadable. Anyone can therefore run it on their own machine, modify it, and build their own service on top of it. By contrast, the flagship models of most large American competitors are available only through the developer's own servers.

How it works

1. The company trains a large language model on vast amounts of text. The model learns what word likely follows another, and this forms the basis of its response.
2. It uses efficient solutions: the model consists of many smaller 'expert' components, and only a few are active when generating each word. Thus, less computing capacity is required.
3. The 'thinking' mode walks through the task step by step before answering and shows you this line of reasoning. Originally, this was possible only with the separate DeepSeek-R1 model; later, the company merged normal and thinking modes, and switching between them is done with a button in the application.
4. You can access the finished model in three ways: as a free chatbot on the web and in the mobile app, through a paid programming interface (API), or by downloading it and running it on your own machine.
5. Since the model is freely downloadable, other companies also integrate it into their products, such as cloud services or coding agents.

An everyday example

Who is affected

A common misconception

Limits and risks

• Do not enter personal, health, or business secrets into the official app, because the data are stored in China, where different laws apply than in the EU.
• On topics sensitive to the Chinese government (such as the Tiananmen Square events or Taiwan), the model evades the question, stays silent, or repeats the official position.
• Like every language model, it can make confident mistakes, so always verify important facts from other sources.
• Authorities in several countries have restricted or investigated the service for data protection reasons, so its availability and perception vary by country.

Similar, but not the same

• ChatGPT – ChatGPT is a product of the US-based OpenAI, and the underlying top-tier models cannot be downloaded. DeepSeek is a Chinese development, and its models are freely downloadable. OpenAI released downloadable models in 2025, but these were not ChatGPT's top-tier models.
• Open-source model – This is a broader category, and DeepSeek is one well-known example. In its case, the trained model is available for download under a free license, but the training data is not public, so it is more accurate to speak of an "open-weight" model.
• Qwen – Qwen is Alibaba's model family, that is, a similarly open but separately developed product from another Chinese company.

Related concepts

A brief history

DeepSeek was founded in 2023 by Liang Wenfeng in Hangzhou, China, with the backing of the High-Flyer investment fund. It gained worldwide fame in January 2025 with the release of the DeepSeek-R1 model. The app topped the US App Store download charts, and Nvidia lost approximately 600 billion dollars in market value on January 27, 2025, marking the largest single-day loss in US stock market history. In the second half of 2025, the company merged normal and reasoning modes into a single model.

Where things stand

By the end of 2026, DeepSeek is one of the best-known Chinese AI players: its chatbot is used by the masses, and developers worldwide build on its open models. Meanwhile, in the West, its use is restricted in several government and corporate environments for data-protection and security reasons.

Status: 2026 November

In the news

• DeepSeek's agent platform reaches 1.54 million daily users as Chinese open models spreadNov 2, 2026 · Agent Age
• DeepSeek's open-source Super-Agent prototype positioned as an alternative to Claude CodeNov 2, 2026 · Code
• DeepSeek unveils proactive coding agent prototype to challenge AnthropicOct 31, 2026 · Code
• DeepSeek unveils coding "super-agent" and open-sources its architectureOct 27, 2026 · Agent Age
• DeepSeek raises about $15 billion ahead of possible IPOOct 21, 2026 · Agent Age
• China approves overseas use of DeepSeek and MiniMax models as DeepSeek open-sources its inference engineOct 19, 2026 · Agent Age
• US bans DeepSeek from federal agencies as Beijing protestsOct 17, 2026 · Agent Age
• DeepSeek to raise at least $12 billion with Tencent backing as Beijing probes data leakageOct 5, 2026 · Agent Age

Also known as

DeepSeek · Deep Seek · DeepSeek AI · DeepSeek-R1 · DeepSeek-V3 · DeepSeek Chat · 深度求索 · Chinese ChatGPT

Sources

• Wikipedia: DeepSeek
• DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (arXiv)
• DeepSeek – hivatalos weboldal

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.