To use DeepSeek, go to chat.deepseek.com or the DeepSeek mobile app, sign in with an email, Google, or Apple account, and start typing — the free chat app runs on the DeepSeek-V4 models with no paid tier required. Two toggles above the input box, DeepThink and Search, control most of what makes it different from a basic chatbot.
Short answer: Go to chat.deepseek.com (or the iOS/Android app), sign in free with email, Google, Apple, or phone, and type your question. Turn on DeepThink for step-by-step reasoning on hard problems, and Search when you need current web results. Upload PDFs, Word docs, CSVs, or images with the paperclip icon. Everything here works with no subscription.
In my testing, I set up a fresh DeepSeek account, ran the same coding and research prompts I use to test every chatbot, and timed how long each toggle actually took to matter. Signing in and sending a first message took under a minute; figuring out when DeepThink was worth the extra wait took a few more prompts. Here's what I found, step by step, plus the mistakes that waste people's time on their first day using it.
What you'll need
Nothing to install if you're on desktop — chat.deepseek.com works in any browser, and DeepSeek's iOS and Android apps are free downloads for phone use. You'll need an email address, or a Google or Apple account, to sign in; phone-number sign-in is available in some regions too. No credit card is required for the chat app itself, since it has no paid subscription tier at all. If you plan to build with the API instead of the chat app, you'll separately need a DeepSeek platform account and a payment method, since API usage bills per token from the first call.
Step-by-step: How to Use DeepSeek
1. Create an account
Go to chat.deepseek.com and click Sign in. Pick one method — email and password, Google, or Apple — and stick with it; signing in with email once and Google later on the same address can split your chat history across two accounts. The whole process takes under a minute and doesn't ask for payment details.
2. Ask your first question
Type into the chat box and press enter. By default you're talking to DeepSeek's current flagship model, and a plain question gets a plain, direct answer — no separate mode-picking required for everyday use.
3. Turn on DeepThink for harder problems
Above the input box, click DeepThink before you send a tricky question — a multi-step math problem, a debugging task, anything where you want to see the reasoning, not just the conclusion. DeepThink shows its chain-of-thought before landing on a final answer, which is genuinely useful for catching where it went wrong versus where it got it right.
4. Turn on Search for anything time-sensitive
The Search toggle sits next to DeepThink and lets DeepSeek pull in live web results instead of answering purely from training data. Flip it on for current prices, recent news, or anything that changes week to week — leave it off for stable, general-knowledge questions to get a faster reply.
5. Upload files with the paperclip icon
Click the paperclip in the bottom-right of the chat box to attach a PDF, Word document, plain text file, CSV, code file, or image. DeepSeek reads the file directly into the conversation, so you can ask it to summarize a contract, debug a script, or describe a screenshot without retyping anything.
6. Keep going with follow-ups
Ask a follow-up in the same thread and DeepSeek keeps the context — no need to re-paste your original question. You can copy code blocks and tables straight out of the response, and search your past chat history from the sidebar if you want to pick up an old conversation.
Example prompts you can copy
A vague prompt gets a vague answer. Naming the format, the constraint, or the mode you want gets something you can actually use:
- "Turn on DeepThink and solve this step by step, showing your work before the final answer: [paste the problem]."
- "Turn on Search and tell me the current price of [product/service], with the source it came from."
- "Write a [language] function that does [task], then explain the time complexity in two sentences."
- "Here's a contract I'm reviewing [attach PDF] — flag any clause that limits liability or auto-renews."
- "Explain [concept] the way you'd explain it to someone who's never touched the topic, then give one concrete example."
Naming DeepThink or Search directly in the prompt is the fastest way to get the mode you actually wanted, instead of hoping the default behavior matches.
Common mistakes to avoid
The one that cost me the most time early on: leaving DeepThink switched on for simple questions. It works, but it's slower and often over-explains something that didn't need a reasoning trace at all — I now only flip it on when I actually want to see the steps. Second, forgetting to turn on Search for anything time-sensitive; in my testing, asking about a recent event with Search off got me a confident answer built on stale training data, with no warning that it might be outdated. Third, dumping one giant, multi-topic PDF into a chat and then asking a narrow question about page two — it works, but a shorter, relevant excerpt gets a tighter answer. Fourth, assuming the API shares the chat app's free allowance; the chat app has no paid tier, but the API bills per token starting with your first request. Fifth, skimming past DeepThink's reasoning trace instead of reading it — the value of seeing the steps is catching a wrong assumption before it becomes a wrong answer.
DeepSeek chat vs. DeepSeek API, at a glance
| Chat app (web + mobile) | Developer API | |
|---|---|---|
| Price | Free — no paid tier | Billed per token from the first call |
| Model choice | Auto-served current flagship model | Choose deepseek-v4-flash or deepseek-v4-pro |
| DeepThink reasoning | One-click toggle in the UI | Enabled via a request parameter |
| Web search | One-click toggle in the UI | Not built in — you wire up your own retrieval |
| File uploads | PDF, Word, plain text, CSV, code, images | Send content as text in the prompt |
| Context window | Same underlying 1M-token models | 1M tokens in, up to 384K tokens out |
| Best for | Everyday chat, quick research, one-off files | Coding pipelines, bulk automation, embedding in your own product |
Confirm current numbers on DeepSeek’s own API pricing page before building anything that depends on them — as of the date I checked, deepseek-v4-flash runs $0.14 per million input tokens on a cache miss and $0.28 per million output tokens, while deepseek-v4-pro runs $0.435 per million input tokens and $0.87 per million output, with cache-hit input priced far lower on both. DeepSeek's chat app itself has never had a subscription tier — the API is the only place any of this shows up on a bill.
Tools that make this easier
DeepSeek isn't the only free-to-start chatbot worth having open, and knowing where it fits next to the others saves you from reaching for the wrong tool. If you're weighing DeepSeek against the most obvious alternative, my full DeepSeek vs. ChatGPT test covers pricing, output quality, and where each one actually wins. If you're new to chatbots generally, how to use ChatGPT and how to use ChatGPT for free cover the equivalent setup on OpenAI's side. For writing and editing work that needs more polish than a quick DeepThink answer, how to use Claude AI is the tool I reach for next. If coding is your main use case, our roundup of ChatGPT alternatives for coding puts DeepSeek's cheap, reasoning-focused models next to the other options. And if you're still deciding what belongs in your toolkit at all, how we rate AI tools and our list of free AI tools are good next stops.
My take
DeepSeek is worth having open the moment you want a free, capable chatbot for coding, math, or quick research, and the DeepThink and Search toggles cover most of what separates a "just okay" answer from a genuinely useful one — you just have to remember to flip them on. It launched in January 2025 and reached the top of the US App Store within weeks, and the V4 releases in April 2026) kept it competitive on reasoning benchmarks without adding a subscription. Where it still falls short of a tool like ChatGPT is breadth — no voice mode, no image generation, and a thinner app ecosystem — so I keep both installed rather than picking one. The habit worth building either way: match the toggle to the question instead of leaving everything on by default.
Frequently Asked Questions
Is DeepSeek free to use?
Yes. The chat app on web, iOS, and Android has no paid subscription tier — signing up costs nothing and every feature covered in this guide, including DeepThink and Search, works on the free account. Only the separate developer API charges per token.
How long does it take to learn how to use DeepSeek?
A couple of minutes for the basics — sign in, type a question, read the answer. Learning when DeepThink and Search are actually worth turning on, rather than leaving both off or both on by habit, usually takes a handful of real questions.
What is the easiest way to get good results from DeepSeek?
Name the mode you want in the prompt ("turn on DeepThink" or "turn on Search"), upload only the file or excerpt that's actually relevant to your question, and read the reasoning trace instead of skimming past it when accuracy matters.
Is DeepSeek safe to use with sensitive documents?
Treat it the way you'd treat any cloud chatbot: fine for general work, not for anything you wouldn't want processed on a third-party server. DeepSeek is a Chinese company, and its terms route data through servers there — a real consideration for regulated or sensitive material regardless of output quality.
Do I need to create an account to use DeepSeek?
Yes, an account is required to save chat history and use file uploads, but signing up is free and takes under a minute with an email, Google, or Apple login. There's no guest mode for the full feature set covered here.