POST/v1/responses

Responses API

Compatible with the OpenAI Responses API format. Compared with Chat Completions, it offers built-in tools (web search, code interpreter, etc.), a more flexible input format, and simplified multi-turn context management. Call it via the OpenAI SDK's client.responses.create().

⚠ Protocol Restriction

The Responses API only supports the Qwen series. Calling models like GLM, DeepSeek, Kimi, or MiniMax returns an Unsupported model error; use /v1/chat/completions or /v1/messages instead.

Currently supported Qwen series models: qwen3.7-max, qwen3.7-plus, qwen3.6-plus, qwen3.5-plus, qwen3.5-flash, qwen3.6-flash, qwen-plus, qwen-flash, qwen3-coder-plus, qwen3-coder-flash, and more.

Request Endpoint

POSThttps://nexusflow.hk/v1/responses

Request Parameters

Parameter	Type	Required	Description
`model`	string	*	Model name, e.g. qwen3.7-plus, qwen3.7-max
`input`	string \| array	*	Plain text or an array of messages (supports role: user/assistant/system/developer)
`instructions`	string	-	System instructions, inserted at the start of the context
`previous_response_id`	string	-	Previous response ID (valid for 7 days), used for multi-turn conversations
`stream`	boolean	-	Whether to enable streaming output, default false
`store`	boolean	-	Whether to store the response (default true); if false, it cannot be referenced via previous_response_id
`tools`	array	-	Tool list: web_search, web_extractor, code_interpreter, function, etc.
`tool_choice`	string \| object	-	Tool selection strategy: auto / none / required
`temperature`	float	-	Sampling temperature, range [0, 2)
`top_p`	float	-	Nucleus sampling probability threshold, range (0, 1]
`enable_thinking`	boolean	-	Whether to enable thinking mode
`reasoning`	object	-	Thinking effort control, e.g. {effort: "high"}

Code Examples

curl https://nexusflow.hk/v1/responses \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.7-plus",
    "input": "Hello!"
  }'

Built-in Tools

Enabling web_search, web_extractor, and code_interpreter together is recommended for best results.

Tool Type	Description
`web_search`	Web search to fetch the latest internet information
`web_extractor`	Web extraction to pull page content (use with web_search)
`code_interpreter`	Code interpreter that runs code and returns results
`web_search_image`	Text-to-image search, finds images from a text description
`image_search`	Image-to-image search, finds similar images from an image
`file_search`	Knowledge base search over uploaded knowledge bases
`function`	Custom function tool; returns function_call when the model decides to call it

Response Format

Non-streaming requests return a complete JSON object whose object field is "response". Streaming requests return an SSE event stream that ends with the response.completed event.

Field	Type	Description
`id`	string	Unique response identifier (UUID format), valid for 7 days, usable as previous_response_id
`object`	string	Always "response"
`status`	string	completed / failed / in_progress / cancelled
`model`	string	The model ID actually used
`output`	array	Array of output items: message / reasoning / function_call / web_search_call, etc.
`usage`	object	Token usage: input_tokens, output_tokens, total_tokens, input_tokens_details, output_tokens_details

Auxiliary Endpoints

Method	Endpoint	Description
GET	`/v1/responses/{id}`	Retrieve a stored response (requires store=true)
DELETE	`/v1/responses/{id}`	Delete a stored response
GET	`/v1/responses/{id}/input_items`	Get the list of input items used to generate the response

Supported Models

Protocol Restriction

The Responses API only supports the Qwen series. Calling models like GLM, DeepSeek, Kimi, or MiniMax returns an Unsupported model error. Use /v1/chat/completions or /v1/messages instead.

The Responses API currently supports the Qwen series: qwen3.7-max, qwen3.7-plus, qwen3.6-plus, qwen3.5-plus, qwen3.5-flash, qwen3.6-flash, qwen-plus, qwen-flash, qwen3-coder-plus, qwen3-coder-flash, and more.

Notes

The response linked by previous_response_id is valid for 7 days.
Enabling the built-in tools together (web_search + web_extractor + code_interpreter) is recommended for best results.
When store: false is set, the response is not stored and cannot be referenced later.
The final response.completed event in streaming output contains the complete usage information.

Chat Completions

OpenAI Chat Completions endpoint

Parameters

Detailed reference for all model parameters

Context Cache

Reduce multi-turn conversation costs