Create a chat completion

Request body · application/json

Type

object

Required fields

modelmessages

Request body required

• modelstringRequired

  Model ID

• messagesarrayRequired

  List of conversation messages

  messages child fields (1)

  • []objectOptional
    [] child fields (6)

    • rolestringRequired

      Message role

      Enum

      ["system","user","assistant","tool","developer"]

    • contentunionRequired

      Message content

      oneOf branch 1

      Type: string
      oneOf branch 2

      Type: array

      • []objectOptional
        [] child fields (6)

        • typestringOptional

          Enum

          ["text","image_url","input_audio","file","video_url"]

        • textstringOptional

        • image_urlunionOptional

          Supports string URL/Base64 or object {url, detail} wire inputs. Available only to models supporting image input; this is not a guarantee that all models support vision.

          oneOf branch 1

          Image URL or Base64 data URL.

          Type: string
          oneOf branch 2

          Object input must provide at least url. Message.ParseContent normalizes omitted detail to high; support depends on the selected vision model.

          Type: object

          Required in this branch: url

          • urlstringRequired

            Image URL or base64

          • detailstringOptional

            Enum

            ["low","high","auto"]

        • input_audioobjectOptional
          input_audio child fields (2)

          • datastringOptional

            Base64-encoded audio data

          • formatstringOptional

            Enum

            ["wav","mp3"]

        • fileobjectOptional
          file child fields (3)

          • filenamestringOptional

          • file_datastringOptional

          • file_idstringOptional

        • video_urlstringOptional

          The actual wire input read by Message.ParseContent is a video URL string, available only to models supporting video input. A {url: ...} object does not enter this parser's content/metering results and is not published as a usable contract.

    • namestringOptional

      Sender name

    • tool_callsarrayOptional
      tool_calls child fields (1)

      • []objectOptional
        [] child fields (3)

        • idstringOptional

        • typestringOptional

        • functionobjectOptional
          function child fields (2)

          • namestringOptional

          • argumentsstringOptional

    • tool_call_idstringOptional

      Tool call ID (for tool-role messages)

    • reasoning_contentstringOptional

      Reasoning content

• temperaturenumberOptional

  Sampling temperature. Specific support and limits depend on the selected model; no uniform defaults are guaranteed.

• top_pnumberOptional

  Nucleus sampling parameter. Specific support and limits depend on the selected model; no uniform defaults are guaranteed.

• nintegerOptional

  Number to generate. Specific support and limits depend on the selected model; no uniform defaults are guaranteed.

• streambooleanOptional

  Whether to stream the response

  Default value

  false

• stream_optionsobjectOptional

  Chat streaming usage options; support and whether usage is returned depend on the selected model.

  stream_options child fields (1)

  • include_usagebooleanOptional

• stopunionOptional

  Stop sequences

  oneOf branch 1

  Type: string
  oneOf branch 2

  Type: array

  • []stringOptional

• max_tokensintegerOptional

  Maximum generated token count. Specific support and limits depend on the selected model; no uniform defaults are guaranteed.

• max_completion_tokensintegerOptional

  Maximum completion token count. Specific support and limits depend on the selected model; no uniform defaults are guaranteed.

• presence_penaltynumberOptional

  Specific support and limits depend on the selected model; no uniform defaults are guaranteed.

• frequency_penaltynumberOptional

  Specific support and limits depend on the selected model; no uniform defaults are guaranteed.

• logit_biasobjectOptional
  logit_bias child fields (1)

  • [key]numberOptional

• userstringOptional

• toolsarrayOptional
  tools child fields (1)

  • []objectOptional
    [] child fields (2)

    • typestringOptional

    • functionobjectOptional
      function child fields (3)

      • namestringOptional

      • descriptionstringOptional

      • parametersobjectOptional

        Parameter definition in JSON Schema format

• tool_choiceunionOptional
  oneOf branch 1

  Type: string

  Enum: ["none","auto","required"]
  oneOf branch 2

  Type: object

  • typestringOptional

  • functionobjectOptional
    function child fields (1)

    • namestringOptional

• response_formatobjectOptional
  response_format child fields (2)

  • typestringOptional

    Enum

    ["text","json_object","json_schema"]

  • json_schemaobjectOptional

    JSON Schema definition

• seedintegerOptional

  Specific support and limits depend on the selected model; no uniform defaults are guaranteed.

• reasoning_effortstringOptional

  Reasoning effort (for models that support reasoning). Specific support and limits depend on the selected model; no uniform defaults are guaranteed.

• modalitiesarrayOptional
  modalities child fields (1)

  • []stringOptional

    Enum

    ["text","audio"]

• audioobjectOptional
  audio child fields (2)

  • voicestringOptional

  • formatstringOptional

• parallel_tool_callsbooleanOptional

  Whether parallel tool calls are allowed depends on the selected model. Chat accepts a boolean; Responses accepts compatible protocol JSON values.

• storebooleanOptional

  Whether requests and responses may be stored depends on the selected model and supported service capabilities. This does not provide a general conversation storage API.

• prompt_cache_keystringOptional

  prompt cache correlation identifier; model dependent, with no guarantee of a cache hit or support. Chat accepts a string; Responses accepts compatible protocol JSON values.

• prompt_cache_retentionunknownOptional

  prompt cache retention control as a JSON value; model dependent. The platform provides no uniform enum, default, or cache retention guarantee.

• service_tierunknownOptional

  Requested service tier, which may affect charges. Ignored by default and effective only when supported. Accepted values depend on the selected model.

HTTP 200

{
  "type": "string",
  "description": "SSE 事件流，协议事件/用量/结束符取当前接口，错误可能在流内。"
}

HTTP 400

HTTP 401

HTTP 403

HTTP 429

HTTP 500