Chat

Responses

Create a model response with the OpenAI Responses API.

POST /v1/responses

Compatible with the OpenAI Responses API — the newer interface used by reasoning models, Codex and agent frameworks. It supports text and image input, built-in and function tools, reasoning settings and multi-turn state via previous_response_id.

Use this endpoint with models whose supported_endpoint_types include openai-response (see List models). For some other providers, such as Gemini, the request is converted to Chat Completions internally; in that case previous_response_id and conversation are not supported. When in doubt, use Chat Completions.

Authentication

Authorization: Bearer sk-xxxx

Request body

Parameter	Type	Required	Description
`model`	string	Yes	Model ID, e.g. `gpt-5`, `gpt-4o-mini`.
`input`	string \| array	Yes	A text prompt, or an array of input items (messages with `input_text` / `input_image` / `input_file` parts, previous outputs, tool results).
`instructions`	string	No	System / developer instructions inserted before the input.
`max_output_tokens`	integer	No	Upper bound for generated tokens, including reasoning tokens.
`temperature`	number	No	Sampling temperature, `0`–`2`.
`top_p`	number	No	Nucleus sampling, `0`–`1`.
`tools`	array	No	Tools the model may use: `function` tools and built-in tools such as `web_search` when the upstream supports them.
`tool_choice`	string \| object	No	`none`, `auto`, `required`, or a specific tool.
`parallel_tool_calls`	boolean	No	Allow several tool calls in one turn.
`max_tool_calls`	integer	No	Maximum number of built-in tool calls.
`reasoning`	object	No	Reasoning settings: `effort` (`minimal`, `low`, `medium`, `high`) and `summary` (`auto`, `concise`, `detailed`).
`text`	object	No	Text output settings. `text.format` selects `{"type": "text"}`, `{"type": "json_object"}` or `{"type": "json_schema", "name": ..., "schema": ..., "strict": true}`.
`previous_response_id`	string	No	Continue a conversation from a previous response stored by the upstream.
`store`	boolean	No	Whether the upstream may store the response (required for `previous_response_id` later).
`include`	array	No	Extra output data, e.g. `["reasoning.encrypted_content"]`.
`truncation`	string	No	`auto` or `disabled`.
`metadata`	object	No	Up to 16 key-value pairs attached to the response.
`stream`	boolean	No	Stream events as SSE. See Streaming.
`prompt_cache_key`	string	No	Key that improves prompt cache hit rates for similar requests.
`user`	string	No	Stable end-user identifier.

background mode is not supported. previous_response_id relies on the upstream provider having stored the earlier response, so it only works if the follow-up reaches the same upstream account. For reliable multi-turn conversations, send the full history in input.

Request example

cURL

curl https://rutoken.ai/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -d '{
    "model": "gpt-5",
    "instructions": "Answer concisely.",
    "input": "Which is larger, 9.11 or 9.9?",
    "reasoning": {"effort": "low"},
    "max_output_tokens": 1000
  }'

Python

from openai import OpenAI

client = OpenAI(base_url="https://rutoken.ai/v1", api_key="sk-xxxx")
resp = client.responses.create(
    model="gpt-5",
    instructions="Answer concisely.",
    input="Which is larger, 9.11 or 9.9?",
    reasoning={"effort": "low"},
    max_output_tokens=1000,
)
print(resp.output_text)

Node.js

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://rutoken.ai/v1", apiKey: "sk-xxxx" });
const resp = await client.responses.create({
  model: "gpt-5",
  instructions: "Answer concisely.",
  input: "Which is larger, 9.11 or 9.9?",
  reasoning: { effort: "low" },
  max_output_tokens: 1000,
});
console.log(resp.output_text);

Image input uses an input_image part:

json

{
  "model": "gpt-4o-mini",
  "input": [
    {
      "role": "user",
      "content": [
        {"type": "input_text", "text": "Describe this image."},
        {"type": "input_image", "image_url": "https://example.com/cat.jpg"}
      ]
    }
  ]
}

Response

json

{
  "id": "resp_67ccd2bed1ec8190",
  "object": "response",
  "created_at": 1760000000,
  "status": "completed",
  "model": "gpt-5",
  "output": [
    {
      "type": "reasoning",
      "id": "rs_67ccd2bf",
      "summary": []
    },
    {
      "type": "message",
      "id": "msg_67ccd2bf",
      "role": "assistant",
      "status": "completed",
      "content": [
        {"type": "output_text", "text": "9.9 is larger.", "annotations": []}
      ]
    }
  ],
  "usage": {
    "input_tokens": 24,
    "output_tokens": 70,
    "total_tokens": 94,
    "input_tokens_details": {"cached_tokens": 0},
    "output_tokens_details": {"reasoning_tokens": 64}
  }
}

Field	Type	Description
`id`	string	Response ID, usable as `previous_response_id`.
`status`	string	`completed`, `incomplete`, `failed` or `in_progress`.
`output`	array	Output items: `message`, `reasoning`, `function_call`, built-in tool calls.
`output[].content[].text`	string	Generated text of a `message` item. The SDKs expose the concatenation as `output_text`.
`usage`	object	`input_tokens`, `output_tokens`, `total_tokens` and their details (cached and reasoning tokens).

Streaming

With "stream": true the endpoint emits typed events such as response.created, response.output_text.delta and response.completed (which carries usage). See Streaming.

Compaction

POST /v1/responses/compact compacts a long conversation into a shorter context that you can pass as input to the next request. It accepts model, input, instructions and previous_response_id, and returns an object with id, output and usage. Availability depends on the upstream channel; it is mainly used by Codex.

bash

curl https://rutoken.ai/v1/responses/compact \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_KEY" \
  -d '{
    "model": "gpt-5-codex",
    "input": [{"role": "user", "content": "...long conversation..."}]
  }'

← Previous

Chat Completions

Claude Messages