Responses

POST /v1/responses provides a Responses-style input shape over the same orchestration and billing engine as Chat Completions.

{
  "model": "gpt-5.6-sol",
  "instructions": "Be concise.",
  "input": "What is a request ID?",
  "max_output_tokens": 128,
  "stream": false
}

The stateless translation is explicit:

Responses field	Canonical behavior
`instructions`	Prepended as a system message
string `input`	One user message
message-array `input`	Passed through as messages
`max_output_tokens`	`max_completion_tokens`
`text.format`	`response_format`
`reasoning.effort`	`reasoning_effort`

Responses output uses object: "response", an assistant message output item, normalized usage, and the inferencepass request and charge metadata. Stateful fields such as store, previous_response_id, conversation, and background are not supported.

Both endpoints use the same admission, capability, routing, reservation, usage, and settlement path. Choose the surface that best fits the client library.