Learn DeepSeek in 5 minutes.
Lesson not loading? Open it in a new tab. It is the same free lesson either way.
Chat, DeepThink, or Search?
The fast default
The standard model replies quickly and handles writing, explanations, and everyday code well. Start here; escalate only when the question resists.
The R1 reasoner
DeepThink routes your question to DeepSeek's reasoning model, which thinks visibly before answering. You can read the chain of reasoning it produced, which is unusual, and useful for checking its logic.
Live web results
The Search toggle grounds answers in current web pages instead of training data. Turn it on for anything dated: news, prices, versions, releases.
What you will do in the lesson
- Break a question on the default. You find a problem the quick model fumbles, so the next toggle has something to prove.
- Rerun it with DeepThink. You put the same question to the reasoning model and compare both the answer and the visible thinking that led there.
- Read the trace, not just the answer. You practise scanning the reasoning for the step where it commits, which is where errors hide.
- Flip on Search. You ask something time-sensitive with and without web access and see how grounding changes the reply.
- Combine the toggles deliberately. You finish by matching a set of tasks to the right combination: chat, think, search, or both.
Before you start
Do I need to install anything or make an account?
Not for this lesson; it runs in the page. DeepSeek itself is free to use in a browser with a sign-up, and the models are also published openly for others to host.
What does open model actually mean here?
DeepSeek releases its model weights publicly, so companies and researchers can run or modify them on their own hardware. That is why you also meet DeepSeek models inside other apps and tools.
Where does my data go if I use the official app?
Conversations on DeepSeek's own service are processed on its servers in China, under its privacy policy. For sensitive work that matters, and many organisations use DeepSeek models through third-party hosts instead. The lesson itself sends nothing to DeepSeek.
I already use ChatGPT. What would DeepSeek add?
The interesting parts are the visible reasoning and the economics: DeepThink shows its work in a way ChatGPT mostly does not, and the open models make the same capability cheap to run anywhere. Try it on a hard reasoning problem and compare.
