# How do you try DeepSeek-R1, and what about privacy?

Date: 2025-01-29 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/try-deepseek-r1-about-privacy/

**Answer:** You can try DeepSeek-R1 for free by turning on the DeepThink option at chat.deepseek.com or in the DeepSeek app, call it through the API as deepseek-reasoner, or download its open weights. DeepSeek's privacy policy says the data it collects, including your prompts, is stored on servers in China, so decide what you type with that in mind.

DeepSeek-R1, released on January 20 by the Chinese lab DeepSeek, has become one of the most talked-about AI models of the month, and its app has climbed the download charts. If you want to see what the fuss is about, there are three ways in: the free chat website and app, the paid API for developers, and the open model files anyone can download. Each one handles your data differently, which is why the privacy question deserves a few minutes before you start.

R1 is a reasoning model: before it answers, it writes out a long chain of thought, working through the problem step by step. DeepSeek says it performs on par with OpenAI's o1 on math, code and reasoning tests.

## Three ways to try it

| Route | How | Cost | Where your prompts go |
|---|---|---|---|
| Chat website or app | chat.deepseek.com or the mobile app; sign in and turn on **DeepThink (R1)** | Free | DeepSeek's servers |
| API | Model name `deepseek-reasoner` on DeepSeek's developer platform | $0.55 per million input tokens ($0.14 when cached), $2.19 per million output tokens | DeepSeek's servers |
| Open weights | Download the model files, released under the MIT license | Your own hardware | Stay on the machine running it |

A token is a small piece of text, roughly three-quarters of an English word, and it is the unit AI companies bill by. The MIT license is one of the most permissive open-source licenses: it allows commercial use and lets others build on the model.

## A worked reasoning test

Reasoning models shine on problems that need several steps and punish a quick guess. Turn on DeepThink and send a puzzle like this:

```prompt
A bat and a ball cost $1.10 together. The bat costs $1.00 more than the ball. How much does the ball cost? Then explain, in two sentences, why many people get this wrong.
```

Before the answer, R1 shows its thinking: it typically tries the intuitive answer of 10 cents, checks it, notices that the bat would then cost $1.10 and the total $1.20, and corrects itself to 5 cents (the description of its reasoning is illustrative; the steps vary from run to run). Reading that visible reasoning is the most instructive part. It shows you where the model checked its own work, and it helps you spot the step where it went wrong when it does.

> **Try it:** Ask R1 the same puzzle with DeepThink off, then on. Compare the answers and count how many times the reasoning version checks or revises itself before replying.

## What the privacy policy says

DeepSeek's privacy policy, last updated December 5, 2024, is a short read and worth doing. Here is what it states, in plain terms:

- **Who runs the service.** Two DeepSeek companies, one based in Hangzhou and one in Beijing, both registered in China.
- **What it collects.** Account details such as email or phone number; everything you type, upload or say to the model, including chat history; and technical data such as your device model, IP address and "keystroke patterns or rhythms."
- **How it's used.** Among other purposes, to review and improve the service, including training and improving its technology.
- **Where it's stored.** On servers in the People's Republic of China.
- **When it's shared.** With advertising and analytics partners, and with law enforcement or public authorities when the company believes in good faith that it is necessary.
- **Your controls.** You can delete your chat history in settings and, depending on where you live, may have rights to access or delete your data.

None of this is unusual on its own: most chat services collect prompts and device data and use them to improve their products. What sets this policy apart for many users, and for regulators, is the storage location and which country's laws apply to the data.

## What regulators have said so far

On January 28, Italy's data protection authority, the Garante, sent DeepSeek's two companies a formal request for information. Citing a possible high risk to the data of millions of people in Italy, it asked what personal data the service collects, from which sources, for what purposes and on what legal basis, whether that data is kept on servers in China, and what information is used to train the model. The companies have 20 days to reply. A request for information is a first step, not a finding that rules were broken.

For everyday use, a sensible approach is the same one you'd apply to any free online service: don't paste in anything you wouldn't want stored by a company abroad, such as health details, client files, passwords or your employer's confidential documents.

## Can you run it on your own computer?

Yes, but not the full model on ordinary hardware. The complete R1 is very large and needs server-class machines with many high-end graphics cards. What most people can run is one of the six smaller "distilled" versions DeepSeek released alongside it, built on open models from the Qwen and Llama families and trained to imitate R1's reasoning. DeepSeek says its 32-billion- and 70-billion-parameter versions match o1-mini on its tests; the smallest versions fit comfortably on a recent laptop.

Running a model locally changes the privacy picture completely. Nothing you type leaves your machine, so the questions above about storage and sharing no longer apply. Free tools such as Ollama and LM Studio make this practical without writing code: you download the model once, then chat with it offline.

The trade-off is quality and speed. A small distilled model reasons less reliably than the full R1 on DeepSeek's servers, and it runs only as fast as your computer allows. For sensitive documents, many people will find that trade worth making; for a quick puzzle or a general question, the website is simpler.

## Sources

- DeepSeek-R1 Release: https://api-docs.deepseek.com/news/news250120
- DeepSeek Privacy Policy: https://chat.deepseek.com/downloads/DeepSeek%20Privacy%20Policy.html
- AI: The Italian Data Protection Authority requests information from DeepSeek: https://www.garanteprivacy.it/home/docweb/-/docweb-display/docweb/10096856