What DeepSeek is
DeepSeek is an AI company in Hangzhou, China, founded in July 2023 by Liang Wenfeng, according to Wikipedia. Its chatbot runs on the website at deepseek.com and in iOS and Android apps, and the models behind it are open-weight, which means the model files are published for anyone to download.
The chatbot became widely known in January 2025. Its R1 model passed ChatGPT as the most downloaded free app on the US iOS App Store on 27 January, Wikipedia reports. On 6 November 2026 the home page lists the V4 family, with V4.1-Flash updated in its 1017 version, and a free chat next to a paid API.
What DeepSeek is good for
- Everyday questions and drafts: answers, emails and summaries, in English or Chinese.
- Step-by-step reasoning: with DeepThink on, the model works through a problem before it answers, which suits maths, logic puzzles and tracing a bug. DeepSeek's docs say thinking mode is meant to improve accuracy.
- Questions that need fresh facts: switch on Search and it looks at the web before it answers.
- Long documents: the V4 models take a context window of one million tokens, enough for a book-length text.
- Code: the chat writes and explains programs. DeepSeek's docs say tools such as Claude Code and GitHub Copilot can use its API as the backend model.
- Your own apps: the API follows the OpenAI and Anthropic formats, so most apps built for those can switch by changing the address and the key.
When to use DeepSeek, and when to pick another
| Your job | Use DeepSeek? | Or try |
|---|---|---|
| A free everyday chatbot for questions, writing and code help | Yes. The web chat and the app are free to use | ChatGPT, Claude or Gemini |
| A hard maths or logic problem where you want the working shown | Yes, with DeepThink switched on | Gemini, ChatGPT or Claude, which also have deeper thinking modes |
| Private files, client data or personal details | No. The official app runs on DeepSeek's servers | Ollama or LM Studio, to run an open model on your own computer |
| Answers about topics that are politically sensitive in China | Not first. Reports say the official app limits some answers | ChatGPT, Claude or Gemini |
| A report with sources from the live web | Partly. Search looks at the web | Perplexity |
| Building your own app on a model API | Yes. The API uses the OpenAI and Anthropic formats | The Claude API or the OpenAI API |
How to use DeepSeek, step by step
These steps follow the web chat, with the switch names as DeepSeek writes them on its home page.
- Open DeepSeek. Go to deepseek.com and select Chat with DeepSeek, or install the DeepSeek app from your phone's app store.
- Sign in. Create an account or log in when the page asks. The chat is free to use.
- Ask a question. Type in the message box and press Enter. Say what the answer is for, such as "a reply to a customer" or "a study guide".
- Switch on DeepThink for hard problems. Select DeepThink before you send. The model works through the problem first, then answers.
- Switch on Search for fresh facts. Select Search and the model looks at the web before it answers.
- Give it your material. Paste a long text into the box. The V4 context window is one million tokens, and a 2026 tutorial by Parker Prompts also covers file analysis.
- Check what matters. For a figure, a date or a legal or health point, confirm it on a source you can open.
On the DeepSeek home page, select Chat with DeepSeek to open the web chat.
Type your question in the Message DeepSeek box and press Enter. Say what the answer is for.
Select DeepThink before you send. The model works through a hard problem first, then answers.
Select Search when you need fresh facts. The model looks at the web before it answers.
AI-made illustrations of DeepSeek's screens, drawn on 6 November 2026 to match these steps. They are not screenshots, so the real app may look different: follow what you see.
The tutorials at the top of this page show these steps on screen.
An example to try
Paste this into the chat with DeepThink switched on. It has one right answer you can work out yourself.
A train leaves at 9:15 and travels 180 km at 72 km/h. It waits 25 minutes, then travels another 90 km at 60 km/h.
At what time does it arrive? Show each step.- The correct arrival time is 13:40, which is 1:40 in the afternoon. The first leg takes 2 hours 30 minutes and the second 1 hour 30 minutes, with 25 minutes of waiting between them.
- Run it with DeepThink off and then on, and compare how each answer lays out the steps.
- Whatever the answer, redo one step by hand. That habit is what keeps you safe on problems you cannot check so easily.
Tips for better results
- Name the form you want: a table, three bullets or a short email.
- Leave DeepThink off for quick rewrites, because it spends time reasoning first.
- Put your text and your question in the same message, with the question last.
Privacy, limits on answers and where it runs
The official chat runs on DeepSeek's servers, so what you type goes there. Regulators have raised concerns about its data practices. Wikipedia lists Italy ordering it to block its chatbot in January 2025, South Korea suspending new downloads in February 2025, and a US order, published on 20 October 2026, banning it from civilian government agencies. Keep passwords, client files and personal details out of the chat.
Reports say the official app limits answers on topics the Chinese government treats as sensitive, such as the Tiananmen Square massacre and the political status of Taiwan. Users who host the open models themselves report removing some of that, though locally run copies still sometimes give answers that match official narratives.
Learn more about DeepSeek
- 2Which AI model should I use for my task?Which AI model should I use? Pick the assistant by your task and where your files live, then a fast or deep-thinking model, and test two on your own work.
- 3Hugging Face Inference API: call hosted open models with Inference ProvidersCall hosted open models through Inference Providers, the successor to the Hugging Face Inference API: a token, InferenceClient, the OpenAI client and curl.
- 4Is it safe to use AI chatbots with personal data? What the apps keepIs it safe to use AI chatbots with personal data? See what the big apps say they keep, how to turn off training, and what never to paste into a chat.
Frequently asked questions
What is DeepSeek?
It is a Chinese AI company, founded in Hangzhou in July 2023, and the free chatbot and open-weight models it makes. The chat runs on its website and in phone apps.
Is DeepSeek free?
The chat on the website and in the app is free to use, Wikipedia says. The API is billed by usage, so read the API pricing page before you build on it.
Is DeepSeek safe to use?
Keep private data out of it. The official chat runs on DeepSeek's servers, and regulators in Italy, South Korea and the United States have acted on its data practices. For general questions, many people use it.
What is DeepThink?
It is a switch in the chat that makes the model reason through a problem before it gives the answer. DeepSeek's docs call this thinking mode, and it is meant to improve accuracy.
Is DeepSeek open source?
Its models are open-weight, so the files can be downloaded, and the V4 models carry the MIT license. The training data is not openly licensed, Wikipedia notes.
Can DeepSeek search the internet?
Yes, when you switch on Search in the chat. The API also supports web search and fetch, according to the home page.
How do I use the DeepSeek API?
Point an OpenAI-compatible app at https://api.deepseek.com, add your API key and set the model to deepseek-flash or deepseek-v4-pro. Those are the names the docs list on 6 November 2026.
Sources
- DeepSeek: home page https://www.deepseek.com/en/
- DeepSeek API Docs: Your First API Call https://api-docs.deepseek.com/
- DeepSeek API Docs: Thinking Mode https://api-docs.deepseek.com/guides/thinking_mode
- Hugging Face: deepseek-ai/DeepSeek-V4-Flash https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash
- Wikipedia: DeepSeek https://en.wikipedia.org/wiki/DeepSeek
- Wikipedia: DeepSeek (AI) https://en.wikipedia.org/wiki/DeepSeek_%28AI%29
- YouTube: Parker Prompts, Learn 97% of DeepSeek AI in 11 Minutes https://www.youtube.com/watch?v=d8XoXf0cSJI
- YouTube: Simplilearn, How To Use DeepSeek? DeepSeek Tutorial For Beginners https://www.youtube.com/watch?v=ybvUzU1JnM4