DeepSeek is the AI chatbot that briefly made it into every news bulletin in early 2025, and it still raises more questions than almost any other tool.
What is it, exactly?
Why is it completely free when rivals charge?
And why have several governments banned it from official devices?
This guide covers the basics in plain language: what DeepSeek actually is, what it genuinely helps with, why it costs nothing, where the data goes, and which restrictions and limits are worth knowing before choosing it. No jargon, no hype, just a clear starting point.
What DeepSeek actually is
DeepSeek is a chatbot made by a Chinese AI company of the same name, based in Hangzhou and founded in July 2023 by Liang Wenfeng. The company is owned and funded by High-Flyer, a Chinese hedge fund Liang co-founded in 2015, whose computing infrastructure supported the AI research that became DeepSeek.
Like ChatGPT, it is built on a large language model: a type of AI trained on huge amounts of text so it can understand questions and generate original responses.
Most people first heard of it in January 2025. On 27 January 2025, the DeepSeek app overtook ChatGPT as the most downloaded free app on Apple's US App Store. The same day, Nvidia's shares fell nearly 17%, erasing close to $600 billion in market value: the largest single-day market value loss for any company in US stock market history.
Markets reacted that way because DeepSeek appeared to match the leading American models at a fraction of the cost. The current flagship is the V4 model family, unveiled on 24 April 2026, and like every chatbot, the model underneath changes regularly.
What DeepSeek can help with
DeepSeek covers the same core tasks as other chatbots: answering questions, drafting and editing text, summarising documents, explaining topics, translation, and coding help. Someone comfortable with the basics of ChatGPT will find the experience familiar.
Its particular reputation is for reasoning, maths, and coding. That reputation is not just marketing. When its R1 reasoning model launched on 20 January 2025, independent reviews found it performed comparably to OpenAI's best reasoning model at the time on maths and coding problems.
The free app includes a reasoning mode, file uploads, and web search. One more distinctive feature: DeepSeek publishes the weights of some of its models, with R1 released under an MIT licence. In plain terms, technically skilled people can download the model and run it on their own hardware.
For most beginners, though, the headline is simpler. Everything above is completely free.
Is DeepSeek free?
Yes, entirely. The web chat and the official mobile apps have no subscription tiers, no Plus or Pro plan, and no paywall as of July 2026. There is no paid consumer version to upgrade to.
That is a genuine difference from rivals, most of which reserve their best models or higher usage limits for paying subscribers.
The only paid product is the developer API, which businesses use to build DeepSeek into their own software, and it is unusually cheap. The official pricing lists DeepSeek-V4-Flash at $0.28 per million output tokens, a fraction of what major rivals charge. Everyday users never touch this; it is simply why so many companies have taken notice.
Free is appealing. But with DeepSeek, the cost shows up somewhere else, and that is the part of this guide that matters most.
DeepSeek privacy: where the data goes
DeepSeek's own privacy policy is unusually direct about this. It states, word for word:
> "To provide you with our services, we directly collect, process and store your Personal Data in People's Republic of China." (DeepSeek privacy policy, last updated 10 February 2026)
What gets collected is broad. Per the same policy, it includes:
- text and voice chats, uploaded files, and chat history
- account details: username, email, phone number, date of birth
- device and network data, including IP address and device identifiers
- approximate location derived from IP address
- usage logs of features used and actions taken
The policy also states data may be shared with service providers, corporate group entities, and "law enforcement and government agencies", and is retained "for as long as necessary to provide our Services".
One point often gets reported wrongly, so it is worth being precise. The policy does state users have the right to opt out of their data being used to train DeepSeek's models. An opt-out exists. What the opt-out does not do is move the data anywhere. Opting out of training still leaves everything stored in China, and regulators have questioned how effectively those user rights work in practice.
Why does the storage location matter? Reporting has consistently noted that companies operating in China can be legally compelled to share data with Chinese authorities, including data on users outside China. That is a legal exposure, not evidence of anything specific happening to any individual's chats. It is also the stated reason behind most of the government action below.
The practical rule is the same one that applies to every AI tool, only more so here. The safest approach is to assume anything typed into DeepSeek is stored in China indefinitely, and to keep personal, financial, and work-confidential information out of it entirely.
Restrictions and bans
A growing list of governments has restricted DeepSeek, mostly on official devices and mostly citing that same data exposure. The verified timeline:
- Italy became the first country to act, when its data protection authority ordered a block on DeepSeek processing Italians' data on 30 January 2025
- Taiwan blocked government departments from using DeepSeek in early February 2025, citing security risks
- Australia banned all DeepSeek products from government systems and devices on 4 February 2025
- South Korea blocked it on government devices, then went further: on 15 February 2025 the app was removed from local app stores pending a privacy review, after DeepSeek acknowledged failing to comply with local data protection rules
- Germany's data protection commissioner has asked Apple and Google to remove DeepSeek from their stores, and the Czech Republic has banned its public administration from using any DeepSeek services on the advice of its national cyber security agency
In the United States, there is no nationwide consumer ban, though some government equipment is restricted and legislation on government devices has been advancing in Congress.
Note the pattern. These are mostly government-device bans, not bans on ordinary people using the app. In most countries, anyone can still download DeepSeek freely. The restrictions are a signal about how governments rate the data risk, not a rule for the general public.
Censorship and limitations
DeepSeek has a limitation no major Western chatbot shares: it refuses to answer, or gives Chinese government positions on, politically sensitive topics. Documented examples include questions about Tiananmen Square in 1989, Taiwan's status, the treatment of Uyghurs, and criticism of Xi Jinping.
The behaviour can be visible in real time. The app has been observed starting to generate an answer about Tiananmen Square and then deleting it mid-response, replacing it with a refusal.
This is not a glitch. Chinese AI regulations require models offered in China to align with state content rules, and the behaviour is trained into the model itself, so running it locally does not remove it.
Most everyday users will never ask about Chinese politics. The censorship still matters, because it shows the tool's answers are shaped by rules beyond accuracy.
On top of that, the normal chatbot limits apply in full. DeepSeek can be confidently wrong, a problem usually called hallucination. Its strength on maths benchmarks does not mean every answer is right. Its output should be treated as a starting point rather than a finished answer, and it should not replace qualified professionals for medical, legal, or financial decisions.
How DeepSeek compares to other AI tools
For everyday writing, explaining, and planning, the experience is closer to its rivals than the headlines suggest. All the major chatbots take a plain-language request and generate a response.
The genuine differences, per 2026 comparison reviews: DeepSeek wins on price and logic-heavy tasks, while ChatGPT offers a bigger ecosystem of features and more polish for general and creative work. Reviewers consistently recommend ChatGPT for general use and DeepSeek for reasoning-heavy or cost-sensitive work.
The bigger difference is not capability at all. It is where the data lives. Anyone weighing the two can read the companion guide, Everything You Need to Know About ChatGPT, for the same honest breakdown of that tool.
Final thoughts
DeepSeek is a genuinely capable chatbot, strong at reasoning and coding, and completely free at a time when rivals charge for their best features. On ability alone, it earns its place among the major AI tools.
The trade-off is equally real. Its own privacy policy states that everything typed into it is stored in China, the opt-out covers training rather than storage, and multiple governments have restricted it on their own devices for that reason. Separately, its answers on certain topics follow rules set by the Chinese state rather than accuracy.
None of that makes it unusable. It makes it a choice that should be made deliberately, with the facts in view, rather than by downloading whatever is top of the app chart.
The practical approach: anyone drawn to DeepSeek for its reasoning strength or its price can use it for exactly that, keep anything personal or confidential out of it completely, and check important answers against a second source. AI works best as a helper, not the final authority, and with DeepSeek the choice of helper carries more weight than usual.
