DeepSeek is a free AI chatbot from China that runs in your browser and on iPhone and Android. In 2026 it changed a lot: new V4 models in April, one unified chat mode in September, and its own search engine in October.
This guide shows you how to sign up, what each button does, how to keep your data private, and which prompts work well. Everything here is checked against DeepSeek’s own pages and app store listings as of 19 November 2026.
- Where: chat.deepseek.com in any browser, or the free “DeepSeek – AI Assistant” app on iOS and Android.
- Price: the chat is free. No paid plan or in-app purchase is listed on the App Store.
- Model: the web and app chat run DeepSeek-V4.1-Flash-1017 since 17 October 2026.
- Modes: Instant, Expert and Vision were merged into one mode in September 2026. The model decides how hard to think.
- Privacy: data is stored in China. Turn off “Help improve the model for everyone” in Settings, Data.
- Developers: use the API (pay per token) or run the open weights on your own hardware.
What DeepSeek Is in November 2026
DeepSeek is made by Hangzhou DeepSeek Artificial Intelligence Co., Ltd. It offers a free chat assistant, a paid developer API, and open model weights that anyone can download.
Most beginners only need the chat. It works like ChatGPT or Claude: you type a question, it answers, and it can read files, look at images and search the web.
| Way to use DeepSeek | Who it is for | Cost |
|---|---|---|
| Web chat (chat.deepseek.com) | Anyone with a browser | Free |
| Mobile app (iOS, Android) | Phone users | Free, no in-app purchases listed |
| API (platform.deepseek.com) | Developers and apps | Pay per token |
| Open weights (Hugging Face) | People with strong hardware | Free download, you pay for hardware |
Here is what changed recently, so older guides do not confuse you:
- 24 April 2026: DeepSeek-V4 (Pro and Flash) launched, and 1M token context became the default on all official services.
- 11 September 2026: app version 2.5.0 merged Instant, Expert and Vision modes into one.
- 17 October 2026: web and app moved to V4.1-Flash-1017, and search switched to DeepSeek’s in-house engine.
- 13 November 2026: app features like search, file reading and voice were opened to developers through the API.
How to Sign Up on the Web and the App
DeepSeek’s privacy policy lists these sign-in options: email or phone number with a password, or Google and Apple login “where available”. The exact buttons depend on your country.
On the web
- Open the official site. Go to chat.deepseek.com. Check the address bar carefully, because many look-alike sites copy the name.
- Choose a sign-in method. Pick email, phone, or Google login if it is shown.
- Verify your account. Enter the code sent to your email or phone, then set a password.
- Accept the terms. The Terms of Use say the service is “primarily intended for adults”. Under-18s should use it with a guardian’s consent.
- Start chatting. Type in the box at the bottom and press Enter.
On iPhone or Android
- Find the right app. Search “DeepSeek – AI Assistant”. On iOS the seller is Hangzhou DeepSeek Artificial Intelligence Co., Ltd. On Google Play the package is com.deepseek.chat.
- Check the requirements. The iOS app needs iOS 15.0 or later and is about 78.5 MB. It is rated 13+ on the App Store and 12+ on Google Play.
- Install and open it. Sign in with the same account you use on the web, so your chats sync.
- Allow only what you need. Grant camera, photo or microphone access only when you want to use those features.
Warning: fake “DeepSeek” apps and websites are common. The official app is free, so never pay or enter card details to “unlock” DeepSeek chat.
Which Model the Chat Uses Now
DeepSeek’s change log for 17 October 2026 says the official web and app models were updated to DeepSeek-V4.1-Flash-1017. That is the same model developers call as deepseek-flash in the API.
V4.1 Flash arrived on 10 September 2026. It is the smallest model in DeepSeek’s new architecture family, and it has native image understanding, so one model now handles text and pictures.
| Model | Where you find it | Notes |
|---|---|---|
| V4.1-Flash-1017 | Web chat, app, API (deepseek-flash) |
1M context, thinking on by default, vision |
| V4-Pro-0813 | API (deepseek-v4-pro) |
1M context, no vision in the API |
| V4 Flash, V4 Flash Vision Exp | Retired | Old names now route to V4.1 Flash |
| deepseek-chat, deepseek-reasoner | Retired 24 July 2026 | Old API names from the V3 era |
Thinking Mode (DeepThink) and Web Search
What happened to DeepThink
Older guides tell you to press a DeepThink button to make the model reason step by step. That button came from the V3 era. In April 2026 it was replaced by Instant and Expert modes.
Then the September app update merged Instant, Expert and Vision into one mode. The model judges how hard your question is and decides how much to think. The app notes also say the thinking process now collapses automatically, so you see a short summary you can expand.
Note: DeepSeek renames these buttons often. If your screen still shows a thinking toggle, turn it on for maths, logic and code, and off for quick chat. Follow the labels in your own app.
Using web search
Since 17 October 2026, search in the web chat runs on DeepSeek’s own search engine. It is useful for news, prices, release dates and anything newer than the model’s training data.
- Turn search on if your screen shows a Search option. Otherwise just ask a question about recent events.
- Ask a clear question. For example, “What changed in the latest iPhone update? List sources.”
- Check the citations. Open two or three of the linked pages before you rely on the answer.
DeepSeek also describes “search while thinking”. With thinking on, the model can search, read pages, and then search again based on what it found. This works well for research questions with several parts.
Files, Images and Voice
Uploading files
Click the paperclip or plus icon next to the message box, then pick a file. DeepSeek’s developer guide lists PDF, Word, PowerPoint and image files as examples of what its app-style document reading supports.
- Good uses: summarise a report, pull a table out of a PDF, or explain a contract clause in plain English.
- Bad uses: payslips, ID scans, medical records, or anything under an NDA.
- Limits: DeepSeek does not publish a file size or count limit for the chat. If an upload fails, split the file.
Asking about images
The app added vision mode in June 2026 and improved photo and file uploads in July. With V4.1 Flash, image understanding is built in. Upload a screenshot, chart or photo and ask your question.
Try “What error is shown in this screenshot and how do I fix it?” or “Turn this whiteboard photo into neat notes”. DeepSeek reads images but does not generate images.
Voice input
The privacy policy lists voice input among the data the app handles, and says DeepSeek does not extract voiceprints. DeepSeek’s 13 November release also calls voice interaction one of its app features.
On a phone, look for the microphone icon in the message box. If you do not see one, your keyboard’s own dictation button works just as well.
Chat History and Data Controls
Your chats are saved to your account and appear in the left sidebar. You can rename or delete single chats from the menu next to each one. The bigger controls live in Settings.
To open them, click the sidebar (top left), click the three dots next to your name (bottom left), and choose Settings. Then use these options:
| Setting | Where | What it does |
|---|---|---|
| Help improve the model for everyone | Settings, Data | Turn off to opt out of training on your chats |
| Shared links | Settings, Data, Manage | Delete public links you created |
| Export data | Settings, Data, Export | Download your chats (link lasts 7 days) |
| Delete all chats | Settings, Data, Delete all | Wipes your whole chat history |
| Delete account | Settings, Profile, Delete | Permanent; you type a confirmation phrase |
Tip: export your data before you delete your account. DeepSeek’s privacy policy says a deleted account cannot be reactivated and its content cannot be recovered.
The exported history is also useful for developers. DeepSeek’s open-source demo chat client, released on 13 November 2026, can import a history exported from DeepSeek Chat.
Limits and Privacy Basics
Usage limits
DeepSeek does not publish a daily message cap for the free chat. The Terms of Use do ban bulk account sign-ups and bots that scrape the service. They also say DeepSeek makes no promise the service will always be available.
- Context: 1M tokens is the default on official services, so long documents usually fit in one chat.
- Busy periods: at peak times you may see a “server busy” message. Wait a minute and retry.
- Topics: the hosted chat avoids some politically sensitive topics about China.
Privacy in plain words
The privacy policy (updated 10 February 2026) says DeepSeek collects, processes and stores personal data in the People’s Republic of China. It collects your prompts, uploads, voice input, chat history, device details and IP-based location.
It keeps data for as long as you have an account. It may use your data to train models, and you have a right to opt out. Some governments ban DeepSeek on official devices. Read our full breakdown in Is DeepSeek safe?
Warning: the Terms of Use say outputs “may contain errors” and are not professional advice. Always check medical, legal and money answers with a qualified person.
10 Prompt Examples to Try
Good prompts give context, a task and a format. Copy these, then change the details in brackets.
- Explain simply: “Explain [compound interest] to a 12-year-old in 5 short bullet points, with one example.”
- Summarise a file: “Summarise this PDF in 8 bullets. Then list 3 questions I should ask the author.”
- Write an email: “Write a polite 120-word email asking my landlord to fix [a leaking tap] by Friday.”
- Study helper: “Quiz me on [photosynthesis] with 10 multiple-choice questions. Show answers only at the end.”
- Debug code: “This Python code throws [error]. Explain the cause, then give the fixed code.” (Paste your code below.)
- Plan a trip: “Plan a 3-day budget trip to [Lisbon] in a table: day, morning, afternoon, evening, rough cost.”
- Research with search: “Using web search, what were the main AI model releases this week? Cite each source.”
- Read a screenshot: “Here is a screenshot of my router settings. Which option turns on WPA3?”
- Compare options: “Compare [two laptops] for a student in a table, then give a one-line recommendation.”
- Hard reasoning: “Solve this step by step and check your answer at the end: [maths or logic puzzle].”
Tip: if an answer is too long, reply “Shorter, 5 bullets max”. DeepSeek keeps the context, so you can refine instead of starting again.
When to Use the API or Run DeepSeek Locally
The free chat is enough for most people. Move on to the API or a local setup only when you have a clear reason.
Use the API when
- You build an app or bot: the API works with OpenAI-style and Anthropic-style requests.
- You automate tasks: batch summaries, coding agents, or tools like Cline and Continue.
- You need app features in code: since 13 November 2026 the API offers web search, file input, OCR, speech-to-text and text-to-speech.
Prices are low. As of 19 November 2026, deepseek-flash costs $0.15 input (cache miss) and $0.60 output per 1M tokens off-peak, and double that at peak hours. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays.
Follow our step-by-step guide on how to get a DeepSeek API key, then see DeepSeek API pricing for the full table.
Run it locally when
- Privacy comes first: nothing leaves your machine, which removes the China data storage concern.
- You work offline: no internet or server-busy messages.
- You want no topic filters: open weights can be run without the hosted service’s restrictions.
DeepSeek publishes its weights on Hugging Face, but the full models are very large and need server-class hardware. Most home users run smaller distilled versions with Ollama or LM Studio. Our guide to running DeepSeek locally covers which size fits your GPU.
Troubleshooting: Common Problems and Fixes
| Problem | Likely cause | Fix |
|---|---|---|
| “Server busy” message | Peak traffic | Wait a minute, then press regenerate. Try off-peak hours. |
| No verification code arrives | Spam filter or SMS delay | Check spam, wait 60 seconds, or try the other method (email or phone). |
| Google or Apple login missing | Not offered in your region | Sign up with email and password instead. |
| File upload fails | File too big or unusual format | Split the PDF, or convert to PDF, Word or an image. |
| Answers look out of date | No search used | Ask it to use web search, or turn search on. |
| “Beyond my current scope” reply | Topic filter on the hosted chat | Rephrase, or use a local model for neutral research. |
| Chats missing on phone | Signed in with a different account | Log out and sign in with the same method you used on the web. |
| Site will not load | Network block or regional ban | Try the app, another network, or check local rules. |
If the model gives a wrong answer, do not just repeat the question. Add the missing detail, paste the source text, or ask it to “show your steps and check each one”.
Frequently Asked Questions
Is DeepSeek free?
Yes. The web chat and the mobile apps are free, and the App Store lists no in-app purchases. Only the developer API is paid, per token.
Is DeepSeek safe to use?
The app is official and widely used, but your data is stored in China and may be used for training unless you opt out. Avoid sharing sensitive details, and read Is DeepSeek safe? for the full picture.
Is DeepSeek better than ChatGPT?
It depends on the task. DeepSeek is free with long context and strong reasoning, while ChatGPT has image generation, more integrations and paid tiers. See DeepSeek vs ChatGPT for a side-by-side comparison.
Which model does DeepSeek chat use right now?
Since 17 October 2026, the web and app chat use DeepSeek-V4.1-Flash-1017. You cannot pick V4 Pro in the chat; it is offered in the API as deepseek-v4-pro.
Can DeepSeek create images?
No. DeepSeek can read and explain images you upload, but it does not generate pictures.
How do I stop DeepSeek training on my chats?
Open Settings, then Data, and turn off “Help improve the model for everyone”. Chats you already shared before switching it off cannot be pulled back.
Do I need the API to use DeepSeek?
No. The API is only for developers who want to connect DeepSeek to their own apps or scripts. For normal chatting, use chat.deepseek.com or the app.
Sources
- DeepSeek API Change Log
- DeepSeek Models and Pricing
- DeepSeek-V4 release note (24 April 2026)
- DeepSeek guide: Build a chatbot (demo client)
- DeepSeek – AI Assistant on the App Store
- DeepSeek – AI Assistant on Google Play
- DeepSeek Privacy Policy
- DeepSeek Terms of Use
- DeleteMe: DeepSeek privacy settings guide



