DeepSeek
DeepSeek drew attention for its low operating cost and for publishing model weights that others can inspect or run. We'll look at the chat app, its reasoning mode, the developer options, and the costs involved.
What is DeepSeek?
DeepSeek is a chat assistant from a Chinese AI research lab of the same name. Like the others in this course, you type a question and get a conversational reply. What set it apart when it launched was a reasoning model, DeepSeek-R1, that performed close to the leading models at the time while costing far less to build and run.
DeepSeek also releases its model weights publicly. Anyone with the right hardware can download and run them directly, which is unusual among the tools covered so far in this course.
Getting started
- Go to chat.deepseek.com.
- Sign up with an email address, phone number, or Google account.
- Type a message in the box at the bottom and press Enter.
- Toggle "DeepThink" on if you want to see its reasoning before the answer.
What DeepSeek can do
DeepSeek's differences show up most clearly in reasoning mode, open weights, and API pricing. The examples below focus on those areas.
Writing & everyday chat
Drafts, summaries, and general questions when you need a quick response.
Visible reasoning (DeepThink)
Turn on DeepThink and it shows its step-by-step working before giving a final answer, useful for checking its logic.
By 4pm the first train has a 60 km head start. The second train gains 30 km/h on it, so it needs 60 ÷ 30 = 2 hours to close the gap...
Math & logic problems
A particular strength: multi-step word problems, proofs, and calculations where showing the work matters.
Coding help
Writing functions, explaining errors, and working through a piece of code line by line.
File analysis
Upload a document and ask questions about its contents.
Attach a file and ask about it in plain language.
Open-weight models
DeepSeek's models are published for anyone to download, not locked behind one app or API.
Low-cost API
Developers building their own apps pay per token, at rates well below most closed competitors.
This is what caused the stir in the AI industry: comparable quality, a fraction of the running cost.
Web search
An optional web search toggle for questions about current events or anything time-sensitive.
Good to know: data and privacy
DeepSeek is a Chinese company, and chats through its hosted app are processed and stored on its own servers by default. If data residency matters for your work, check your organization's policy first. Because the model weights are open, another option is running DeepSeek yourself, through a local tool like Ollama or a third-party host, so your data never leaves infrastructure you control.
Pricing
DeepSeek adjusts pricing occasionally, so check the official pricing page for current numbers. The main choices are:
- · Generous daily message limits
- · DeepThink reasoning mode
- · File uploads and web search
- · Built for developers, not casual chat
- · Priced well below most closed models
- · Same models used in the chat app
- · No per-message cost once running
- · Full control over where data lives
- · Needs enough GPU memory to run it
DeepSeek vs. ChatGPT and Claude
Day to day, DeepSeek feels similar to the others: a chat box, a message, a reply. The real difference shows up underneath. ChatGPT and Claude are closed models you can only use through their makers' apps and APIs. DeepSeek publishes its weights, so it can also run on hardware you control, not just DeepSeek's own servers.
For everyday writing and general questions, any of them will do the job. DeepSeek is is especially worth testing on math-heavy or logic-heavy problems, or when request cost matters for something you're building.
When DeepSeek is worth trying
DeepSeek is the fifth stop in this AI Tools Course, after ChatGPT, Claude, Gemini, and Perplexity. Up next is Copilot, then Grok and Meta AI, followed by app builders, image tools, and video/audio tools.
What's Next
Next up: Copilot, Microsoft's assistant built into Windows, Edge, and Microsoft 365.