API REFERENCE

API reference

Drop-in OpenAI Chat Completions. Change two lines of config and your existing code runs unchanged.

BASE URL https://apikey.help/v1 MODEL deepseek-v4.1-flash-1017 AUTH Bearer

01Quickstart

Three steps: get a key, change the base URL, send a request. Any OpenAI SDK that accepts a custom base URL works as-is.

  1. Create an API key Sign in and hit "New key" in the console. Copy the string starting with sk-. The full key is shown only once, at creation.
  2. Swap the base URL Replace https://api.openai.com/v1 with https://apikey.help/v1. Nothing else changes.
  3. Send a request Use the model name deepseek-v4.1-flash-1017. Both reasoning and visible output bill at the output rate.
from openai import OpenAI

client = OpenAI(
    base_url="https://apikey.help/v1",
    api_key="sk-your-key",
)

resp = client.chat.completions.create(
    model="deepseek-v4.1-flash-1017",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

02Endpoints

GET/v1/models

List the models available to your key. Handy for checking that a key is valid.

POST/v1/chat/completions

Chat completion. Request and response match OpenAI exactly, including the usage object.

{
  "id": "resp_01a1e90c...",
  "object": "chat.completion",
  "model": "deepseek-v4.1-flash",
  "choices": [{
    "index": 0,
    "message": { "role": "assistant", "content": "Hi! How can I help you today?" },
    "finish_reason": "stop"
  }],
  "usage": { "prompt_tokens": 56, "completion_tokens": 11, "total_tokens": 67 }
}
Note: the model field in the response reads deepseek-v4.1-flash (the upstream alias), which differs from the -1017 suffix you send. Do not match on this field.
Parameter Type Description
modelstring Required. Only deepseek-v4.1-flash-1017 is available. Anything else returns 503 model_not_found.
messagesarray Required. A standard OpenAI message array. system / user / assistant are supported.
streamboolean Optional, defaults to false. Set true to receive an SSE stream.
max_tokensinteger Optional. Capped at 384000.
temperaturenumber Optional. Standard OpenAI semantics.
reasoning_effortstring Optional, but ignored — thinking mode is always on.

03Streaming

Send stream: true to receive text/event-stream. Each line starts with data:, and the stream ends with data: [DONE]. Same as OpenAI.

data: {"choices":[{"delta":{"role":"assistant"},"index":0}]}

data: {"choices":[{"delta":{"content":"Hi"},"index":0}]}

data: [DONE]

04Errors

Every error has the same shape: {"error": {"message", "type", "code"}}. The codes below were verified against the live endpoint.

401
Invalid key Invalid token The key is mistyped, deleted, or the Authorization header is missing. It must be "Bearer sk-xxx", not a bare "sk-xxx".
401
Insufficient balance insufficient quota Check the balance in the console. Top-ups apply instantly; you do not need to recreate the key.
429
Too many requests rate limit There is a cap on concurrent requests. Back off and retry on 429 — do not resend immediately.
503
Model not found model_not_found The model name is wrong. Only deepseek-v4.1-flash-1017 is available; confirm with GET /v1/models first.
500
Upstream error upstream error The upstream returned an error. These calls are never billed, so retrying is free.

05Billing

Billed per token, with input and output priced separately, per million tokens. Failed requests are free.

Item USD / 1M Official list
Inputcache miss
— $0.150
Outputincluding reasoning
— $0.600
Prices come from the live config and match the home page. Balance is denominated in USD; 1 USDC of top-up credits 1 USD.
Minimum top-up 1 USDC on Base. Converted to USD credit automatically once it lands.
Failures are free Upstream errors, timeouts and broken streams never reach your bill.
Thinking mode Always on. Reasoning tokens bill at the output rate and cannot be disabled.

06FAQ

Do LangChain / Cursor / Cline work?

Yes. They all accept a custom base URL and key. Point them at https://apikey.help/v1 — no code changes.

Can I turn thinking mode off?

No. It stays on so answer quality is consistent. Parameters like reasoning_effort are ignored rather than rejected.

How large is the context window?

1M tokens of context, with up to 384K tokens per response.

How do I top up?

Connect a wallet in the console and send USDC on Base. Credit lands automatically after on-chain confirmation.

← Back to home Start free