How to Get a DeepSeek API Key (V3 and R1)
DeepSeek is the reason the phrase cheap reasoning exists. R1 does the long chain of thought that used to be reserved for the most expensive models, at a price that changes what you are willing to point it at.
| Models you get | DeepSeek V3, DeepSeek R1 (Reasoner) |
|---|---|
| Key starts with | sk- |
| What it costs | Prepaid Prepaid balance, topped up in small amounts. Among the cheapest per token anywhere. |
| Time to a working key | about three minutes |
| Console | platform.deepseek.com/api_keys |
Getting the key
- Open platform.deepseek.com/api_keys and sign up with an email address or a Google account.
- Go to Top up and add a small balance. DeepSeek is prepaid, and the minimum is low enough that a first top up covers a lot of work.
- Back on the API keys page, press Create API key and name it.
- Copy it. It starts with
sk-, the same shape as an OpenAI key, so label it carefully if you hold both. - In Calik AI, open API keys, choose DeepSeek, paste and save.
Which model to start on
V3 for everyday work: fast, cheap, competitive with far more expensive general models. R1 when the problem needs an actual chain of reasoning, which is worth the extra tokens for maths, debugging and multi step planning and is wasted on a summary.
When the key does not work
DeepSeek's own wording, and what each one is actually telling you.
| What you see | What it means | What to do |
|---|---|---|
402 Insufficient Balance | The prepaid balance is empty. | Top up on the platform. There is no monthly invoice to fall back on. |
401 Authentication Fails | Mistyped key, or an OpenAI key pasted into the DeepSeek slot: both start with sk-. | Check which provider the key came from before assuming it is broken. |
429 Rate limit reached | Too many concurrent requests. | Reduce concurrency. DeepSeek throttles on parallel requests more than on totals. |
Empty reasoning_content | You asked R1 for a plain answer through a client that hides the reasoning channel. | This is expected: the reasoning stream is separate from the answer stream. |
If you would rather not pay
DeepSeek has no free tier of its own, but its open weights are served free by others: several DeepSeek models appear on OpenRouter's free list, and R1 distills run on Groq and locally under Ollama. The Get API page has all 27 providers with the steps for each.
Where the key goes
Calik AI never resells tokens, so the key you just created is the whole billing relationship: your prompts go from your device straight to DeepSeek, and the bill is theirs to send. The key is written to your device's encrypted storage and is not copied to our servers, our database or our backups.
Open the workspace, find API keys in settings, choose DeepSeek, paste and save. The model picker then lists only what your connected providers actually serve, so there is nothing in it you cannot run.
One key, then the whole workspace
48 agents, 43 AI Center modules and Agent Mesh, all driven by keys that belong to you. Free tier, no card, no expiry.
Create your account See all 27 providersQuestions people ask first
Is the DeepSeek API free?
No, it is prepaid, but it is cheap enough that a few dollars lasts a long time. If you want DeepSeek at no cost, OpenRouter carries free variants of several DeepSeek models and Ollama runs the distilled ones locally.
Is DeepSeek R1 worth it over V3?
For maths, debugging, and any plan with several dependent steps, yes: R1 reasons before it answers and the difference shows. For summaries, rewriting and extraction it burns tokens on thinking you will never read. Keep both connected and switch per task.
Is my data safe with DeepSeek?
That is a question about DeepSeek's terms, not about Calik AI. Your key is stored encrypted on your device and your prompts go straight from you to DeepSeek without passing through us, but once they arrive, DeepSeek's own terms and jurisdiction apply. Read them before sending anything you would not send in a postcard.
Why does my OpenAI key not work in the DeepSeek slot?
Both providers issue keys that start with sk-, which makes them easy to mix up, but they are not interchangeable. Paste the key from the DeepSeek platform into the DeepSeek slot.
Can I compare DeepSeek against GPT on my own question?
Yes, and it is the fastest way to find out whether the cheaper model is good enough for your work. Connect both and let Agent Mesh run them side by side. There is a written comparison too.