DeepSeek
A chat assistant with strong reasoning, free to use in the browser, and an API priced far below the Western labs for anyone building on top of it.
What you'd use it for
Running a workflow that calls a model thousands of times a day — bulk classification, tagging, or summarising — for a fraction of what the same volume costs on a Western API.
- Chat and reason through everyday work for free in a browser or mobile app, no subscription
- Call the API from any OpenAI- or Anthropic-compatible client by swapping the base URL, so existing code keeps working
- Run high-volume automation such as classifying, tagging, or summarising at API prices roughly a tenth of the comparable Western models
- Cut costs further with cached-input pricing and off-peak hours, 01:00-04:00 and 06:00-10:00 UTC on weekdays
- Self-host the open-weight models on your own infrastructure if sending data to China is not acceptable
The verdict
DeepSeek earns its place on cost alone: at roughly a tenth of the comparable Western API prices, it changes the arithmetic for any workflow that calls a model thousands of times over. The chat app is a capable everyday assistant, free, with nothing to configure. What it costs you is trust in where your prompts go — this is a Chinese company processing data on Chinese infrastructure, with no published support tier behind the free chat product. It suits an operator with genuine volume who has settled the data question, not the person making a handful of calls a day where habit already points somewhere else.
What it costs
| Tier | Price | For | Includes |
|---|---|---|---|
| Chat | $0 — the web and mobile assistant is free to use | Anyone who wants a capable assistant without a subscription. | Everyday chat and reasoning in the browser and app. |
| API — deepseek-flash | $0.15/M input tokens off-peak, $0.30 peak; $0.60/M output off-peak, $1.20 peak | High-volume, cost-sensitive automation. | The fast model. Cached input drops to $0.003/M off-peak, $0.006 peak. |
| API — deepseek-v4-pro | $0.66/M input tokens off-peak, $1.32 peak; $1.98/M output off-peak, $3.96 peak | Work where reasoning quality matters more than cost. | The stronger model. Cached input drops to $0.022/M off-peak, $0.044 peak. |
Read from DeepSeek's own pricing page on 25 Sep 2026. Vendors change prices without notice.
What you need to start
- A DeepSeek Platform account and API key for any programmatic use — the chat app itself needs only an email, Google, or Apple sign-in
- A funded, topped-up balance before API calls actually execute — DeepSeek's own billing documentation confirms fees are deducted from that balance
- Comfort routing prompts through infrastructure in China, or the technical capacity to self-host the open-weight models instead
A couple of minutes to a working chat account; longer for the API, since a funded balance has to exist before the first call runs, so treat sign-up and billing as two separate steps.
Your first run Sign-up for the chat app takes an email, Google, or Apple account and you are chatting immediately. For the API, the getting-started page's own curl, Python, and Node examples return a working chat completion within minutes once you have a key and a funded balance — the barrier is billing setup, not the code.
Where it breaks
DeepSeek is a Chinese company and the hosted service processes data in China, which is a governance question your business has to answer before a data question. The consumer chat app carries no published subscription tier, so there is no support relationship to buy. API prices vary by time of day, which makes forecasting a monthly bill harder than it looks. And the pricing page prices tokens, not conversations — a figure that is meaningless until you have measured your own usage.
Browser and mobile apps for chat; a standard HTTP API for building. Off-peak rates are half the peak rates, with peak hours running 01:00-04:00 and 06:00-10:00 UTC on weekdays. The models are also published openly, so they can be self-hosted or run through a third-party provider if the hosted service is not an option for you.
First entry in this reference. There is no previous check to compare it against.
When not to use it
Not for a business handling client or regulated data that has not settled the China-data-residency question — that call needs a lawyer, not this page. Not for someone making a handful of calls a day, where the price difference is invisible and an established provider's support and ecosystem count for more. Not for anyone who wants a paid tier with a support relationship behind the chat app — none is published.
The interesting thing here is not the chat app. It is the price of the API, which is roughly an order of magnitude below the comparable Western models and has been since DeepSeek arrived.
For a business owner, that matters in exactly one situation: when something in your workflow needs a model called thousands of times rather than dozens. Classifying a year of support emails, summarising every inbound enquiry, tagging a product catalogue — work where the per-call cost stops being a rounding error and starts being the budget.
The decision this actually is
It is not “is DeepSeek as good as ChatGPT”. For most one-at-a-time work, the difference in answer quality is smaller than the difference in what you are used to, and habit wins. The decision is whether you have volume, and whether your data can go where this service sends it.
That second half is the one people skip. DeepSeek is a Chinese company processing your prompts on Chinese infrastructure. For a marketing brainstorm that is nobody’s problem; for client records it may be a conversation with your lawyer. The models are also published openly, so a technical team can run them on infrastructure you control — which converts the governance question into a hosting cost.
Where it fits
As a browser tab, it is free and capable and asks nothing of you. As an engine behind an automation in Make or n8n, it is the cheapest serious option on the board.
Alternatives
Assistants & chatAppsChatGPT
OpenAI's assistant in a browser tab, a desktop app and a phone app, where you ask questions, draft text, and hand it files or spreadsheets to work through.
Assistants & chatAppsGoogle Gemini
Google's assistant, usable on its own and also inside Gmail, Docs, Sheets and Drive, so it can draft and summarise against files you already keep in Google.
Assistants & chatModelOpenRouter
A single account and bill for calling hundreds of AI models from different providers, so switching models doesn't mean signing a new contract.
Automation & workflowsHarnessMake.com
A hosted place to build workflows that run on somebody else's servers — triggered by an event, on a schedule, or by a webhook — so the work happens whether or not your laptop is open.
Get DeepSeek
Buying the tool is not the hard part.
The ebook is about what to hand over, in what order, and how you find out when it goes wrong. One payment, no subscription.