DeepSeek in n8n: an API key you can actually pay for

The n8n part is easy. Getting a DeepSeek API key you can pay for is the part that stops most people — so that is what this page is about.

Real cost Free credits Pay with crypto Top-up not working How to pay Payment methods n8n + DeepSeek

The two problems, kept separate

Wiring DeepSeek into n8n takes about two minutes, because DeepSeek speaks the OpenAI format and n8n already knows how to talk to OpenAI-format endpoints. This part is a solved problem and there are a hundred tutorials on it.

Getting the API key is the part that fails. DeepSeek's own platform expects a Chinese phone number for registration and a Chinese payment method for top-ups, which is a dead end for most of the world. So the useful thing this page can tell you is how to get a working key and a base URL, and then the n8n side is genuinely two fields.

Getting a key you can pay for

We're a gateway that holds the upstream relationship and resells access over an OpenAI-compatible endpoint. That means the n8n config is the standard OpenAI one, with a different base URL.

Create an account at api.lbase.com/register. Email and password, no card. You get $2 of free credit — enough to run a few hundred node executions and see whether the quality and latency suit your workflow before you pay anything.
Create an API key in the console. Pick a group when you do. A key with no group returns `API Key is not assigned to any group` on every single request, and the error reads like the gateway is broken. It isn't — it's one required field.
Point n8n at `https://api.lbase.com/v1` and use `deepseek-flash` or `deepseek-v4-pro` as the model name.
When the free credit runs out, top up from $10 with PayPal or USDT (TRC20).

Where the base URL goes in n8n

In n8n's OpenAI credential there's a Base URL field — set it to `https://api.lbase.com/v1`, paste the key, and the OpenAI Chat Model node will talk to DeepSeek. Some older n8n versions hide that field; if yours does, the HTTP Request node always works, since all you need is a POST to `/chat/completions` with a bearer token.

We don't run n8n and we don't maintain an n8n node, so if a specific node version ignores the base URL, believe the node over this page and fall back to HTTP Request. The docs have the raw request shape and configs for Cursor, Cline, Aider, Continue, Dify, LangChain and Claude Code.

A JSON body that works for a plain HTTP Request node:

```json

{

"model": "deepseek-flash",

"messages": [

{ "role": "user", "content": "summarise this: {{ $json.text }}" }

"stream": false

}

```

With header `Authorization: Bearer sk-...` and `Content-Type: application/json`.

The gotchas that actually bite

The group thing. Covered above, but it accounts for a large share of "the API is down" reports. If every request fails identically, check the key before you check the network.
Streaming inside a workflow. If you enable `"stream": true` in a node that expects a single JSON response, you'll get a long silence and then a parse error. Turn streaming off for workflow nodes; it's for chat UIs.
Token growth in loops. Every iteration that re-sends conversation history costs more than the last one. If a workflow loops over a growing context, cost grows roughly quadratically — cap the history you pass in, or summarise it between steps.
Model names change. DeepSeek renamed things across the V4 and V4.1 generations, and old names are tolerated for a while and then aren't. If you hardcoded a model string in twenty nodes, that's twenty places to fix.

What it costs

`deepseek-flash` is $0.55 per million input tokens and $1.65 per million output tokens; `deepseek-v4-pro` is $1.65 / $4.95. DeepSeek applies peak-hours pricing during Chinese business hours (Beijing 09:00–12:00 and 14:00–18:00, Monday to Friday), where token prices double — so a scheduled workflow that runs at 03:00 UTC costs half what the same workflow costs at 03:00 Beijing time. If your jobs are schedulable, that's free money.

The full breakdown, including why the minimum top-up and the payment fee matter more than the per-token rate, is on the pricing page.

The honest part

We are a reseller, not DeepSeek, running on a single node in Hong Kong — treat uptime as best-effort and keep a fallback configured for anything that matters. If you have a direct payment route to DeepSeek, use it; it's cheaper, and the only thing we sell is the ability to pay. We don't store prompt contents beyond what billing and abuse prevention require.

Start free: create an account at api.lbase.com/register — email and password, $2 of credit, no card and no Chinese phone number needed.