mingyan

OpenAI Responses API

Unified response endpoint using the OpenAI Responses API format

POST/gw_llm_power/v1/responses

baseUrlhttp://apillm.globalaiopc.com/gw_llm_power

Compatible with the OpenAI Responses API. You can integrate directly with the following reference:

OpenAI Responses API: https://platform.openai.com/docs/api-reference/responses/create

Authentication

Get Key

Use the Authorization header for authentication.

Required headers:

cURL

Authorization: Bearer {{YOUR_API_KEY}}
Content-Type: application/json

Notes:

Authorization: API key in the format Bearer {{YOUR_API_KEY}}

Supported Models

The Responses API currently supports the GPT models listed in the OpenAI Chat API:

GPT

Official Version	Official Version	Premium Account Pool	Mixed Account Pool / Proxy Version	Low-cost Version
`gpt-5.4`	`gpt-5.4-official`	`gpt-5.4-hc`	`gpt-5.4-fd`	`gpt-5.4-low`
`gpt-5.5`	`gpt-5.5-official`	`gpt-5.5-hc`	`gpt-5.5-fd`	`gpt-5.5-low`

Version Notes

Suffix / Type	Description
no suffix	Official version
`-official`	Official version
`-hc`	Premium account pool version
`-fd`	Account pool / proxy version
`-low`	Low-cost version

Request Parameters

modelstringrequired

Model name to use, for example: gpt-5.4, gpt-5.5, gpt-5.5-official

inputstring/arrayrequired

Model input. Can be a plain text string or a message array using the OpenAI Responses API format

instructionsstring

Developer or system-level instructions that define model behavior, role, and output requirements

previous_response_idstring

Previous response ID used to continue a conversation

temperaturenumber

Sampling temperature to control output randomness. Recommended range is 0 to 2

top_pnumber

Nucleus sampling parameter. It is recommended to tune either temperature or top_p, not both

streamboolean

Whether to enable streaming responses. When set to true, the response is returned as SSE

max_output_tokensinteger

Maximum output token count

toolsarray

Tool definitions available to the model, such as function tools, web search, or file search

tool_choicestring/object

Tool selection policy. The model can choose tools automatically, or you can force a specific tool

textobject

Text output configuration, including structured output formats such as JSON Schema

reasoningobject

Reasoning configuration. When supported by the selected model, this can set reasoning effort and related options

Input Array Structure

input[].rolestringrequired

Message role. Supported values: user, assistant, system, developer

input[].contentstring/arrayrequired

Message content. Can be a plain text string or a multimodal content array

input[].content[].typestringrequired

Content type. Common values include input_text, input_image, input_file, and output_text

input[].content[].textstring

Text content used when type is input_text or output_text

input[].content[].image_urlstring

Image URL used when type is input_image

input[].content[].file_idstring

File ID used when type is input_file

Response

Returns an OpenAI Responses API-compatible response format.

idstring

Unique response identifier. It can be used as previous_response_id

objectstring

Object type, usually response

created_atnumber

Response creation timestamp

statusstring

Response status, such as completed, failed, or in_progress

modelstring

Actual model name used

outputarray

Output array, which can contain messages, tool calls, and other output items

output[].typestring

Output item type, such as message or function_call

output[].content[].typestring

Output content type, such as output_text

output[].content[].textstring

Generated text content

output_textstring

Aggregated text output for direct plain-text reading

usage.input_tokensinteger

Input token count

usage.output_tokensinteger

Output token count

usage.total_tokensinteger

Total token count

Notes

The request endpoint is http://apillm.globalaiopc.com/gw_llm_power/v1/responses
Pass the API key using the Authorization header
Responses API uses input, not the Chat Completions messages field
Streaming responses use SSE events following the OpenAI Responses API format