Inference

Responses

Beta

A text-only subset of the OpenAI Responses API, with streaming.

Endpoint

POST/v1/responses

Requires an API key with the inference scope. Receipts record the operation as inference.responses.

Request body

Parameter	Type	Required	Description
model	string	Required	Model ID from GET /v1/models.
input	string \| array	Required	A non-empty string, or 1–500 input items. See Input formats.
instructions	string	Optional	Up to 200,000 characters. Sent to the model as a system message ahead of the input.
max_output_tokens	integer	Optional	1–128,000. Reduced to the model's cap. Determines the balance reservation.
temperature	number	Optional	0–2. Ignored for models that do not support sampling parameters.
top_p	number	Optional	0–1. Ignored for models that do not support sampling parameters.
stream	boolean	Optional	Stream typed events. Default false.
store	boolean	Optional	Accepted and ignored. Responses are never stored.

Other fields, including tools and previous_response_id, are ignored.

Input formats

A string input becomes a single user message. An array contains message items with a role and content; type: "message" is optional.

Field	Accepted values
role	system, developer (treated as system), user, assistant
content	A string, or up to 64 parts of type input_text, output_text or text. Part texts are concatenated.

Array input

{
  "model": "gpt-5-mini",
  "input": [
    { "role": "developer", "content": "Be brief." },
    { "role": "user", "content": [{ "type": "input_text", "text": "Name three ledger entry types." }] },
    { "role": "assistant", "content": [{ "type": "output_text", "text": "credit, debit, reserve" }] },
    { "role": "user", "content": "And the other two?" }
  ]
}

Any other part type, including images and files, fails validation with HTTP 400.

Example

curl https://www.hushcompute.xyz/v1/responses \
  -H "Authorization: Bearer $HUSH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5-mini",
    "instructions": "Reply in plain text.",
    "input": "Summarize the difference between a budget and a balance.",
    "max_output_tokens": 400
  }'

Response object

200 OK

{
  "id": "resp_01K58ZQ4W5V8E2R7PKD3N6A0TG",
  "object": "response",
  "created_at": 1789387200,
  "status": "completed",
  "model": "gpt-5-mini",
  "output": [
    {
      "type": "message",
      "id": "msg_01K58ZQ4W5V8E2R7PKD3N6A0TG",
      "status": "completed",
      "role": "assistant",
      "content": [{ "type": "output_text", "text": "A balance is ...", "annotations": [] }]
    }
  ],
  "output_text": "A balance is ...",
  "usage": { "input_tokens": 41, "output_tokens": 118, "total_tokens": 159 },
  "receipt": {
    "id": "rcpt_01K58ZQ4W6N3T2B9XRC7D1M5HJ",
    "hash": "0x6f0d…a3c9",
    "signature": "0x9b1f…c07e1b",
    "signer": "0x3A4c…91De",
    "url": "https://www.hushcompute.xyz/proofs/rcpt_01K58ZQ4W6N3T2B9XRC7D1M5HJ"
  },
  "billing": {
    "request_id": "req_01K58ZQ4W5V8E2R7PKD3N6A0TG",
    "cost_micro_usd": "272",
    "cost_usd": "0.000272",
    "usage_estimated": false,
    "fingerprint_salt": "0x51c2…7f08"
  }
}

Field	Type	Description
id	string	resp_ followed by the request ID's suffix. The request ID is in billing.request_id and x-request-id.
status	string	completed, or incomplete when output stopped at max_output_tokens.
output	array	One assistant message with a single output_text part.
output_text	string	Convenience copy of the generated text.
usage	object	input_tokens, output_tokens, total_tokens as billed.
receipt	object	Same shape as Chat Completions: id, hash, signature, signer, url.
billing	object	request_id, cost_micro_usd, cost_usd, usage_estimated, fingerprint_salt.

With stream: true the response is a stream of named Server-Sent Events. Each event has an event: line and a JSON data: line whose type matches the event name. There is no [DONE] sentinel; the stream closes after the last event.

Event	Payload
response.created	response with id, status in_progress and model.
response.output_text.delta	output_index 0, content_index 0, delta with the next text fragment.
response.completed	response: the full response object, including receipt and billing.
error	error object in the standard shape. Sent instead of response.completed.

Wire format

event: response.created
data: {"type":"response.created","response":{"id":"resp_01K5…","status":"in_progress","model":"gpt-5-mini"}}

event: response.output_text.delta
data: {"type":"response.output_text.delta","output_index":0,"content_index":0,"delta":"A balance is"}

event: response.completed
data: {"type":"response.completed","response":{"id":"resp_01K5…","object":"response","status":"completed", … ,"receipt":{…},"billing":{…}}}

Partial-output rules match Chat Completions: if text was delivered before a failure, that portion is billed and receipted with status failed. The error event does not repeat the receipt; its ID is the x-receipt-id header sent when the stream opened. If nothing was delivered, nothing is charged and no receipt exists.

Limitations

No tools, function calling, web search or file search.
No stored state: store is ignored and previous_response_id is not supported.
Text input and output only.
Reasoning items are not returned.

For tool calling use Chat Completions.