Bottom line: DeepSeek is a Chinese AI lab whose current DeepSeek-V4 model family is free to use in its consumer app, published as MIT-licensed open weights, and priced far below US frontier labs through its API. The models are genuinely competitive at coding, reasoning and agentic work, so the practical decision is not about capability but about data governance: the hosted service stores user data on servers in China.
| What it is | An AI research lab in Hangzhou, funded by the hedge fund High-Flyer and led by Liang Wenfeng. It builds the DeepSeek-V4 family (V4-Flash, V4-Pro and an experimental vision model) and ships them three ways: a free chatbot, a paid API, and downloadable weights on Hugging Face. |
| Best for | Cost-sensitive coding, reasoning and agentic workloads; teams that want frontier-adjacent capability they can self-host, fine-tune or run through a provider of their own choosing. |
| Weakest at | Data governance on the hosted service. DeepSeek’s own privacy policy says user data is stored in China, and several governments have blocked or restricted the app on official devices. The hosted models also deflect on topics politically sensitive in China, and capacity is variable enough that the free app returns server-busy errors under load. |
| Pricing | Consumer app and website free, with no paid tier. API billed per token at a small fraction of US frontier-model rates, with discounted off-peak hours. Model weights free to download and run under an MIT licence. |
| Our take | Judge the weights and the hosted service separately. Self-hosted or run through a third-party host, V4 is one of the best capability-per-dollar options available. On the official app, treat anything you type as data leaving your jurisdiction. |
Independent review. No affiliate links. Last checked 23 August 2026.
DeepSeek (Chinese: 深度求索) is a Chinese artificial intelligence company which develops open-source large language models. It is owned and solely funded by Chinese hedge fund High-Flyer, both founded and run by Liang Wenfeng and based in Hangzhou, Zhejiang.
So far, DeepSeek is focused solely on research and has no detailed plans for commercialization.
Frequently asked questions
Is DeepSeek free?
Yes for personal use. The DeepSeek chatbot on the web and in the iOS and Android apps is free, with no paid consumer subscription, although it can slow down or return server-busy messages under heavy load. Developers pay per token for the API, which is priced well below comparable US models. The published model weights are also free to download.
Is DeepSeek safe to use?
The app works as described and is not malware. The real question is data handling: DeepSeek’s privacy policy states that user data is stored on servers in China, and several governments, among them Italy, Australia, Taiwan and South Korea, have blocked or restricted it on official devices. For sensitive or regulated work, run the open weights yourself or use a non-Chinese host.
Who owns DeepSeek?
DeepSeek is a private Chinese company based in Hangzhou, founded in 2023 and funded by the quantitative hedge fund High-Flyer. Both were founded by Liang Wenfeng, who still runs DeepSeek. It has no US parent, no conventional venture-capital syndicate behind it, and it is not publicly traded.
Is DeepSeek open source?
It is open weight rather than fully open source. DeepSeek publishes downloadable model weights on Hugging Face under an MIT licence, which permits commercial use, modification and redistribution. What it does not publish is the training data or the full training pipeline, so you can run and fine-tune the models but cannot reproduce them from scratch.
Is DeepSeek better than ChatGPT?
On published benchmarks the current DeepSeek V4 models are competitive with leading Western models on coding, reasoning and agentic tasks, at a fraction of the API cost. ChatGPT still leads on ecosystem: apps, integrations, voice, image tools and enterprise controls. If cost or self-hosting matters most, DeepSeek wins; if breadth of product matters most, ChatGPT does.
Where is my data stored when I use DeepSeek?
If you use the official app or website, DeepSeek’s privacy policy says your inputs are stored on servers located in China and may be used to improve its models. Chinese law can compel domestic companies to assist state intelligence requests. Running the open weights on your own infrastructure avoids the question entirely, because nothing is sent to DeepSeek.
Which DeepSeek model is the latest?
As of August 2026 the current family is DeepSeek-V4, in two main variants: V4-Flash, the smaller and cheaper model, and V4-Pro, the higher-capability one, which reached general availability on 13 August 2026. An experimental multimodal version, V4-Flash-Vision-Exp, followed on 21 August 2026. These replace the older V3 and R1 lines.