OpenAI-compatible API

Chat Completions

Use POST /v1/chat/completions, with the same semantics as OpenAI. Authenticate with Authorization: Bearer <API Key>.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.goodputai.cn/v1",
    api_key="YOUR_GOODPUT_API_KEY",
)

resp = client.chat.completions.create(
    model="deepseek/deepseek-v4-pro",
    messages=[{"role": "user", "content": "Hello"}],
)

Supported parameters

Parameter	Description
`model`	Required; a model ID from the catalog
`messages`	Required; supports `system` / `user` / `assistant` / `tool` roles
`stream` / `stream_options`	See streaming
`tools` / `tool_choice`	See tool calls
`response_format`	See structured output
`max_tokens` / `max_completion_tokens`	Output limit, including reasoning tokens
`temperature` / `top_p` / `stop`	Sampling and stop controls
`fallbacks`	Platform extension for fallback models; see platform concepts

Parameters unsupported by an upstream model are safely ignored by the gateway without failing the request.

Response structure

Responses follow the OpenAI structure. Reasoning models also return reasoning content and a token breakdown:

{
  "choices": [{
    "message": {
      "role": "assistant",
      "content": "ok",
      "reasoning_content": "The user asked for only ok…"
    }
  }],
  "usage": {
    "prompt_tokens": 17,
    "completion_tokens": 25,
    "total_tokens": 42,
    "completion_tokens_details": {"reasoning_tokens": 22, "text_tokens": 3},
    "prompt_tokens_details": {"cached_tokens": 0}
  }
}

message.reasoning_content: model reasoning; see reasoning models.
usage.completion_tokens_details.reasoning_tokens: reasoning tokens, billed as output.
usage.prompt_tokens_details.cached_tokens: cache-hit tokens, billed at the cache-read price; see prompt caching.

Responses API

POST /v1/responses supports the OpenAI Responses protocol for all platform models:

resp = client.responses.create(
    model="z-ai/glm-5.2",
    input="Introduce yourself in one sentence",
)
print(resp.output_text)

Reasoning is returned in output items with type: "reasoning", or reasoning_summary_text.delta events when streaming. max_output_tokens includes reasoning tokens. For reasoning models, use at least 512; a very small limit can produce status: "incomplete" with an empty answer.

Other endpoints

Endpoint	Description
`GET /v1/models`	Models available to the current key
`POST /v1/completions`	Legacy text completion, retained for compatibility; use Chat Completions for new integrations
`POST /v1/videos`	Submit an asynchronous video generation task
`GET /v1/videos/{video_id}`	Query video task status
`GET /v1/videos/{video_id}/content`	Download the generated video

Notes

A missing or unavailable model returns 404 (code: model_not_found); see errors and rate limits.
Default per-request timeouts start at 300 seconds. Use streaming for long tasks.