OpenAI API Access Guide: Chat Completions Compatibility & Key Configuration

I. Overview

This site provides compatibility with the OpenAI Chat Completions (/v1/chat/completions) format. For code already running with the official OpenAI SDK or OpenAI-compatible clients, simply point your request URL to our site's OpenAI gateway, replace the api_key with your sk- key from our platform, and keep all other call methods the same.

Current available GPT series models are listed in the Model Square (e.g., gpt-5.4, gpt-5.5, gpt-5.4-mini, gpt-4o, etc.). For a complete list and unit pricing, please query GET /v1/models.

II. API Endpoints

Method	Path	Description
POST	/v1/chat/completions	Chat completions (non-streaming / streaming); request body follows OpenAI payload.
GET	/v1/models	List of models and unit prices (no authentication required).

Gateway URL: https://www.relay-api.com. For OpenAI-compatible calls, please send requests to POST https://www.relay-api.com/v1, keeping the original path.

III. Authentication

Authorization: Bearer sk-your_api_key

Create a key starting with sk- in "User Center → API Keys". Use the same authentication header as the official OpenAI API; the SDK will automatically add the Authorization header.

IV. Quick Start

4.1 curl (Non-streaming)

curl -X POST https://www.relay-api.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-your_api_key" \
  -d '{
    "model": "gpt-5.4",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Introduce yourself in three sentences."}
    ],
    "temperature": 0.7,
    "max_tokens": 512
  }'

4.2 OpenAI SDK (Python)

from openai import OpenAI

client = OpenAI(
    base_url="https://www.relay-api.com/v1",
    api_key="sk-your_api_key",
)

resp = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

After setting base_url to the OpenAI entry point above and replacing the api_key, all other invocation methods are identical to the official SDK.

V. Request Parameters

Field	Type	Required	Description
model	string	Yes	Model slug, e.g., gpt-5.4
messages	array	Yes	Message list, elements are {role, content}; role values: system / user / assistant
temperature	number	No	Sampling temperature, default 1.0
top_p	number	No	Nucleus sampling, default 1.0
max_tokens	int	No	Maximum output tokens
stream	bool	No	Enable streaming response, default false
stop	string / array	No	Stop sequence
presence_penalty / frequency_penalty	number	No	Repetition penalty, range -2 to 2

VI. Response Structure

{
  "id": "chatcmpl-123456",
  "object": "chat.completion",
  "created": 1710000000,
  "model": "gpt-5.4",
  "choices": [
    {
      "index": 0,
      "message": {"role": "assistant", "content": "Hello! Glad to meet you."},
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 18,
    "completion_tokens": 12,
    "total_tokens": 30
  }
}

Field	Description
id	Unique ID for this completion
choices[].message	Assistant response; content contains the text
choices[].finish_reason	Termination reason: stop / length / content_filter
usage.prompt_tokens	Input tokens (billing basis)
usage.completion_tokens	Output tokens (billing basis)
usage.total_tokens	Total tokens used

VII. Streaming Output

Adding "stream": true to the request body returns an SSE stream, where each data: line is a chunk:

data: {"id": "chatcmpl-123456", "object": "chat.completion.chunk", "choices": [{"index": 0, "delta": {"role": "assistant"}, "finish_reason": null}]}

data: {"id": "chatcmpl-123456", "object": "chat.completion.chunk", "choices": [{"index": 0, "delta": {"content": "Hello"}, "finish_reason": null}]}

data: {"id": "chatcmpl-123456", "object": "chat.completion.chunk", "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}]}

data: [DONE]

Clients can concatenate choices[0].delta.content from each chunk to get the full response; receiving [DONE] indicates the end of the stream. The OpenAI SDK handles these events automatically when stream=True is passed.

VIII. Common Error Codes

HTTP Status	Meaning	Advice
400	Invalid request parameters (model missing, empty messages, etc.)	Check request body fields
401	Invalid or missing API key	Check Authorization header and key format
403	Key deactivated, insufficient balance, or unauthorized model/route	Top up or check key permissions
404	Path or model not found	Confirm base_url and model name
429	Rate limit triggered	Retry with exponential backoff
500	Internal server error	Retry later; contact support if the issue persists

Error response bodies consistently follow the structure: {"code": status_code, "message": "error_description", "data": null}.

IX. Compatibility Notes

• Supports OpenAI SDKs (Python / Node.js, etc.) via custom base_url.
• Supports various OpenAI-compatible clients (LobeChat, ChatBox, NextChat, One API, etc.) by configuring the custom API URL.
• Model names must match the slugs returned by GET /v1/models.