Chat va completion

This page brings together the OpenAI Chat Completions, OpenAI Responses, native Anthropic Claude, and native Google Gemini chat endpoints.

Pham vi Apifox

OpenAI / ChatCompletions: basic chat, streaming, image understanding, function calling, logprobs, classic completions, web search, PDF analysis, Codex, response_format, N-test variants, token usage
OpenAI / Responses: basic text responses, image analysis, web search tool, file search tool, computer use, streaming, deep-research, reasoning controls, xAI web search
Anthropic Claude: native format, OpenAI-compatible format, forced thinking output, multi-turn tool use, web search
Google Gemini: native chat/media recognition plus OpenAI-compatible reasoning controls, web search, TTS, audio understanding, and video understanding

API model / Chat / OpenAI / Dinh dang Chat Completions

POST Completions Legacy Format

POST /completions

Body Parameters

{
    "model": "gpt-3.5-turbo-instruct",
    "prompt": "Write a tagline for an ice cream shop."
}

Request Parameters

Name	In	Type	Required	Description
Content-Type	header	string	No	none
Accept	header	string	No	none
Authorization	header	string	No	none
body	body	object	No	none
» messages	body	[object]	Yes	Messages used to generate a chat completion in the chat format.
»» role	body	string	No	none
»» content	body	string	No	none
» model	body	string	Yes	Model ID to use. See the model endpoint compatibility table for details on which models work with the Chat API.
» store	body	boolean¦null	No	Whether to store the output of this chat-completion request for distillation or evaluation products.
» reasoning_effort	body	string¦null	No	Available only for `o`-series models.
» metadata	body	object¦null	No	Set of 16 key-value pairs that can be attached to the object. Useful for storing additional structured metadata that can later be queried through the API or dashboard.
» modalities	body	[string]¦null	No	Output types you want the model to generate for this request. Most models produce text, so the default is `["text"]`.
» prediction	body	object	No	Prediction-output configuration. This can greatly improve response time when most of the model response is already known in advance, such as when you are making only small edits to a file.
» audio	body	object¦null	No	Parameters for audio output. Required when requesting audio output with `modalities: ["audio"]`.
» temperature	body	number¦null	No	Sampling temperature between `0` and `2`. Higher values such as `0.8` make output more random, while lower values such as `0.2` make it more focused and deterministic. In most cases, adjust this or `top_p`, but not both.
» top_p	body	number¦null	No	Alternative to temperature sampling, known as nucleus sampling, where the model considers only token results that together make up the `top_p` probability mass. For example, `0.1` means only the top 10% probability mass is considered.
» n	body	integer¦null	No	Number of chat-completion choices to generate for each input message. You are billed for the tokens generated across all choices, so keeping `n` at `1` minimizes cost.
» stop	body	string¦null	No	Not supported by the latest reasoning models or by `.o3` and `o4-mini`.
» max_tokens	body	integer¦null	No	Maximum number of tokens that can be generated in the chat completion. Use this to control text-generation cost through the API.
» max_completion_tokens	body	integer¦null	No	Upper bound on the number of tokens that can be generated in the completion, including visible output tokens and reasoning tokens.
» presence_penalty	body	number¦null	No	A number between `-2.0` and `2.0`. Positive values penalize new tokens based on whether they have already appeared in the text so far, increasing the chance that the model moves to new topics.
» frequency_penalty	body	number¦null	No	A number between `-2.0` and `2.0`. Positive values penalize new tokens based on their existing frequency in the text so far, reducing the chance of repeated lines.
» logit_bias	body	null	No	Adjusts the likelihood that specified tokens appear in the completion.
» logprobs	body	boolean¦null	No	Whether to return log probabilities for output tokens. If set to `true`, the response includes the log probability for each output token in `message.content`.
» user	body	string	No	Unique identifier representing the end user. Helps OpenAI monitor and detect abuse.
» service_tier	body	string¦null	No	Specifies the latency tier used to process the request. This parameter is relevant for customers subscribed to the scale-tier service:
» stream_options	body	object¦null	No	Options for streaming responses. Used only when `stream: true`.
»» include_usage	body	boolean	No	If set, the stream sends an additional block before the `data: [DONE]` message. The `usage` field on that block contains token-usage statistics for the whole request, while `choices` is always an empty array. All other blocks also include `usage`, but with a `null` value. If the stream is interrupted, you might not receive the final usage block.
» response_format	body	object	No	Specifies the format the model must output.
» seed	body	integer¦null	No	Beta feature. If provided, the system makes a best effort to sample deterministically so repeated requests with the same `seed` and parameters should return the same result. Determinism is not guaranteed. Use `system_fingerprint` in the response to monitor backend changes.
» tools	body	[string]	No	List of tools the model may call. Only functions are currently supported as tools. Use this parameter to provide the list of functions for which the model may generate JSON inputs. Up to 128 functions are supported.
» functions	body	[string]	No	Note: deprecated, recommend using tools
» tool_choice	body	string	No	Controls which tool the model calls, if any: `none` means the model generates a message without calling tools; `auto` lets the model choose between generating a message or calling one or more tools; `required` forces one or more tool calls; `{"type": "function", "function": {"name": "my_function"}}` forces a specific tool call.
» function_call	body	string	No	Note: deprecated, recommend using tool_choice
» parallel_tool_calls	body	boolean	No	Whether to enable parallel function calls while tools are being used.
» stream	body	boolean	No	If set to `true`, model response data is streamed to the client as it is generated through server-sent events. See the streaming-response section below and the streaming guide for handling streamed events.
» top_logprobs	body	integer	No	Integer from `0` to `20` specifying how many of the most likely tokens to return at each token position, each with its associated log probability. Requires `logprobs: true`.
» web_search_options	body	object	No	This tool searches the web for relevant results to support the response. See the web-search tool documentation for more details.

Details

» reasoning_effort: Available only for o-series models. Constrains the amount of reasoning effort used by reasoning models. Supported values are low, medium, and high. Lower reasoning effort can improve response latency and reduce the number of reasoning tokens in the response.

» metadata: Set of 16 key-value pairs that can be attached to the object. Useful for storing additional structured metadata that can later be queried through the API or dashboard.

Keys must be strings up to 64 characters long. Values must be strings up to 512 characters long.

» modalities: Output types you want the model to generate for this request. Most models produce text, so the default is ["text"].

This model can also generate audio. To request both text and audio in the same response, use ["text", "audio"].

» top_p: Alternative to temperature sampling, known as nucleus sampling, where the model considers only token results that together make up the top_p probability mass. For example, 0.1 means only the top 10% probability mass is considered.

In most cases, adjust this value or temperature, but not both.

» stop: Not supported by the latest reasoning models or by .o3 and o4-mini. Up to four sequences where the API should stop generating more tokens. The returned text does not include the stop sequence.

» max_tokens: Maximum number of tokens that can be generated in the chat completion. Use this to control text-generation cost through the API.

This field is now deprecated in favor of max_completion_tokens and is incompatible with the o1 model family.

» logit_bias: Adjusts the likelihood that specified tokens appear in the completion.

Accepts a JSON object that maps token IDs to bias values from -100 to 100. Mathematically, the bias is added to the logits generated by the model before sampling. The exact effect varies by model, but values between -1 and 1 should slightly decrease or increase selection likelihood, while values like -100 or 100 should effectively ban or force the relevant token.

» service_tier: Specifies the latency tier used to process the request. This parameter is relevant for customers subscribed to the scale-tier service:

If set to 'auto' and the project has Scale tier enabled, the system uses Scale-tier credits until they are exhausted. If set to 'auto' and the project does not have Scale tier enabled, the request is processed on the default service tier with a lower uptime SLA and no latency guarantee. If set to 'default', the request is processed on the default service tier with a lower uptime SLA and no latency guarantee. If set to 'flex', the request is processed on the Flex Processing service tier. See the vendor documentation for details. If not set, the default behavior is auto. When this parameter is set, the response body includes the service_tier that was used.

» response_format: Specifies the format the model must output.

Set to { "type": "json_schema", "json_schema": {...} } to enable structured output and require the model to match the JSON schema you provide. Set to { "type": "json_object" } to enable JSON mode and require the model to produce valid JSON. Important: when using JSON mode, you must still explicitly instruct the model through a system or user message to produce JSON. Otherwise, the model may emit endless whitespace until the token limit is reached.

» tool_choice: Controls which tool the model calls, if any: none means the model generates a message without calling tools; auto lets the model choose between generating a message or calling one or more tools; required forces one or more tool calls; {"type": "function", "function": {"name": "my_function"}} forces a specific tool call.

Defaults to none when no tools are present, and auto when tools are available.

» function_call: Note: deprecated, recommend using tool_choice Controls which function the model calls, if any:

none: the model will not call a function and will generate a message instead auto: the model can choose between generating a message and calling a function {"name": "my_function"}: forces the model to call a specific function

Response Example

200 Response

{
  "id": "string",
  "object": "string",
  "created": 0,
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "string",
        "content": "string"
      },
      "finish_reason": "string"
    }
  ],
  "usage": {
    "prompt_tokens": 0,
    "completion_tokens": 0,
    "total_tokens": 0
  }
}

Responses

Status	Meaning	Description	Model
200	OK	none	Inline

Response Schema

Status 200

Name	Type	Required	Constraints	Description
» id	string	true	none	none
» object	string	true	none	none
» created	integer	true	none	none
» choices	[object]	true	none	none
»» index	integer	false	none	none
»» message	object	false	none	none
»»» role	string	true	none	none
»»» content	string	true	none	none
»» finish_reason	string	false	none	none
» usage	object	true	none	none
»» prompt_tokens	integer	true	none	none
»» completion_tokens	integer	true	none	none
»» total_tokens	integer	true	none	none

GET Token Usage Query

GET /api/usage/token

Queries the currently authenticated Bearer token for granted quota, used quota, remaining quota, unlimited status, model limits, and expiration time.

Field description (data) object: always token_usage name: token name total_granted: Total granted quota (= used + remaining) total_used: Used quota total_available: Remaining available quota unlimited_quota: Whether quota is unlimited model_limits: Allowed model list model_limits_enabled: Whether model limits are enabled expires_at: Expiration time as a Unix timestamp in seconds. Returns 0 if it never expires (the backend normalizes -1 to 0).

Request Parameters

Name	In	Type	Required	Description
Authorization	header	string	No	none

Response Example

200 Response

{
  "data": [
    {
      "id": "string",
      "object": "string",
      "owned_by": "string",
      "permission": [
        "string"
      ]
    }
  ],
  "object": "string"
}

Responses

Status	Meaning	Description	Model
200	OK	none	Inline

Response Schema

Status 200

Name	Type	Required	Constraints	Description
» data	[object]	true	none	none
»» id	string	true	none	none
»» object	string	true	none	none
»» owned_by	string	true	none	none
»» permission	[oneOf]	true	none	none

oneOf

Name	Type	Required	Constraints	Chinese Name	Description
»»» anonymous	string	false	none		none

xor

Name	Type	Required	Constraints	Chinese Name	Description
»»» anonymous	integer	false	none		none

xor

Name	Type	Required	Constraints	Chinese Name	Description
»»» anonymous	boolean	false	none		none

xor

Name	Type	Required	Constraints	Chinese Name	Description
»»» anonymous	array	false	none		none

xor

Name	Type	Required	Constraints	Chinese Name	Description
»»» anonymous	object	false	none		none

xor

Name	Type	Required	Constraints	Chinese Name	Description
»»» anonymous	number	false	none		none

continued

Name	Type	Required	Constraints	Chinese Name	Description
» object	string	true	none		none

API model / Chat / OpenAI / Dinh dang Responses

POST xAI Web Search

POST /responses

Body Parameters

{
    "model": "grok-4-1-fast-reasoning",
    "input": [
        {
            "role": "user",
            "content": "最近 xAI 有哪些最新动态？"
        }
    ],
    "tools": [
        {
            "type": "web_search"
        },
        {
            "type": "x_search"
        }
    ]
}

Request Parameters

Name	In	Type	Required	Description
Content-Type	header	string	Yes	none
Accept	header	string	Yes	none
Authorization	header	string	No	none
body	body	object	No	none
» model	body	string	Yes	ID of the model to use. See the model endpoint compatibility table for which models are compatible with the Chat API.
» messages	body	[object]	Yes	Messages used to generate a chat completion in the chat format.
»» role	body	string	No	none
»» content	body	string	No	none
» temperature	body	integer	No	Sampling temperature between `0` and `2`. Higher values such as `0.8` make output more random, while lower values such as `0.2` make it more focused and deterministic. In most cases, adjust this or `top_p`, but not both.
» top_p	body	integer	No	Alternative to temperature sampling, known as nucleus sampling, where the model considers only tokens that together make up the `top_p` probability mass. For example, `0.1` means only the top 10% probability mass is considered. In most cases, adjust this or `temperature`, but not both.
» n	body	integer	No	Number of chat-completion choices to generate for each input message.
» stream	body	boolean	No	If set, partial message deltas are streamed the same way they are in ChatGPT. As tokens become available, they are sent as data-only server-sent events, with the stream terminated by `data: [DONE]`. See the OpenAI Cookbook example code.
» stop	body	string	No	Up to four sequences where the API should stop generating additional tokens.
» max_tokens	body	integer	No	Maximum number of tokens to generate for the chat completion. The combined length of input and generated tokens is still limited by the model context window.
» presence_penalty	body	number	No	A number between `-2.0` and `2.0`. Positive values penalize new tokens based on whether they have already appeared in the text so far, increasing the chance that the model discusses new topics. Read more about frequency and presence penalties.
» frequency_penalty	body	number	No	A number between `-2.0` and `2.0`. Positive values penalize tokens based on how often they already appear in the text, reducing the chance that the model repeats the same line verbatim. Read more about frequency and presence penalties.
» logit_bias	body	null	No	Adjusts the likelihood that specified tokens appear in the completion. Accepts a JSON object that maps token IDs to bias values from `-100` to `100`. Mathematically, the bias is added to the logits produced by the model before sampling. The exact effect varies by model, but values between `-1` and `1` should slightly decrease or increase selection likelihood, while values like `-100` or `100` should effectively ban or force the relevant token.
» user	body	string	No	Unique identifier representing your end user. Helps OpenAI monitor and detect abuse. Learn more.

Response Example

200 Response

{
  "id": "string",
  "object": "string",
  "created": 0,
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "string",
        "content": "string"
      },
      "finish_reason": "string"
    }
  ],
  "usage": {
    "prompt_tokens": 0,
    "completion_tokens": 0,
    "total_tokens": 0
  }
}

Responses

Status	Meaning	Description	Model
200	OK	none	Inline

Response Schema

Status 200

Name	Type	Required	Constraints	Description
» id	string	true	none	none
» object	string	true	none	none
» created	integer	true	none	none
» choices	[object]	true	none	none
»» index	integer	false	none	none
»» message	object	false	none	none
»»» role	string	true	none	none
»»» content	string	true	none	none
»» finish_reason	string	false	none	none
» usage	object	true	none	none
»» prompt_tokens	integer	true	none	none
»» completion_tokens	integer	true	none	none
»» total_tokens	integer	true	none	none

API model / Chat / Anthropic Claude / Dinh dang Claude goc (khuyen dung)

POST Web Search

POST /messages

Supports the official native format, using the same model names as the upstream provider

Body Parameters

{
    "model": "claude-sonnet-4-6",
    "max_tokens": 1024,
    "messages": [
        {
            "role": "user",
            "content": "今天几号？用一句话总结今天北京的天气，并给出信息来源"
        }
    ],
    "tools": [
        {
            "type": "web_search_20250305",
            "name": "web_search",
            "max_uses": 3
        }
    ],
    "tool_choice": {
        "type": "auto"
    }
}

Request Parameters

Name	In	Type	Required	Description
Content-Type	header	string	Yes	none
Accept	header	string	Yes	none
x-api-key	header	string	No	none
body	body	object	No	none
» model	body	string	Yes	ID of the model to use. See the model endpoint compatibility table for which models are compatible with the Chat API.
» messages	body	[object]	Yes	Messages used to generate a chat completion in the chat format.
»» role	body	string	No	none
»» content	body	string	No	none
» temperature	body	integer	No	Sampling temperature between `0` and `2`. Higher values such as `0.8` make output more random, while lower values such as `0.2` make it more focused and deterministic. In most cases, adjust this or `top_p`, but not both.
» top_p	body	integer	No	Alternative to temperature sampling, known as nucleus sampling, where the model considers only tokens that together make up the `top_p` probability mass. For example, `0.1` means only the top 10% probability mass is considered. In most cases, adjust this or `temperature`, but not both.
» n	body	integer	No	Number of chat-completion choices to generate for each input message.
» stream	body	boolean	No	If set, partial message deltas are streamed the same way they are in ChatGPT. As tokens become available, they are sent as data-only server-sent events, with the stream terminated by `data: [DONE]`. See the OpenAI Cookbook example code.
» stop	body	string	No	Up to four sequences where the API should stop generating additional tokens.
» max_tokens	body	integer	No	Maximum number of tokens to generate for the chat completion. The combined length of input and generated tokens is still limited by the model context window.
» presence_penalty	body	number	No	A number between `-2.0` and `2.0`. Positive values penalize new tokens based on whether they have already appeared in the text so far, increasing the chance that the model discusses new topics. Read more about frequency and presence penalties.
» frequency_penalty	body	number	No	A number between `-2.0` and `2.0`. Positive values penalize tokens based on how often they already appear in the text, reducing the chance that the model repeats the same line verbatim. Read more about frequency and presence penalties.
» logit_bias	body	null	No	Adjusts the likelihood that specified tokens appear in the completion. Accepts a JSON object that maps token IDs to bias values from `-100` to `100`. Mathematically, the bias is added to the logits produced by the model before sampling. The exact effect varies by model, but values between `-1` and `1` should slightly decrease or increase selection likelihood, while values like `-100` or `100` should effectively ban or force the relevant token.
» user	body	string	No	Unique identifier representing your end user. Helps OpenAI monitor and detect abuse. Learn more.

Response Example

200 Response

{
  "id": "string",
  "object": "string",
  "created": 0,
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "string",
        "content": "string"
      },
      "finish_reason": "string"
    }
  ],
  "usage": {
    "prompt_tokens": 0,
    "completion_tokens": 0,
    "total_tokens": 0
  }
}

Responses

Status	Meaning	Description	Model
200	OK	none	Inline

Response Schema

Status 200

Name	Type	Required	Constraints	Description
» id	string	true	none	none
» object	string	true	none	none
» created	integer	true	none	none
» choices	[object]	true	none	none
»» index	integer	false	none	none
»» message	object	false	none	none
»»» role	string	true	none	none
»»» content	string	true	none	none
»» finish_reason	string	false	none	none
» usage	object	true	none	none
»» prompt_tokens	integer	true	none	none
»» completion_tokens	integer	true	none	none
»» total_tokens	integer	true	none	none

API model / Chat / Google Gemini / Dinh dang Gemini goc

POST Gemini Text Chat

POST /v1beta/models/gemini-3-flash-preview:generateContent

Proxies Gemini API requests. Path format: /v1beta/models/{model_name}:{action} For example: /v1beta/models/gemini-2.5-pro:generateContent /v1beta/models/gemini-2.5-pro:streamGenerateContent?alt=sse

Body Parameters

{
  "contents": [
    {
      "parts": [
        {
          "text": "Hello"
        }
      ]
    }
  ]
}

Request Parameters

Name	In	Type	Required	Description
key	query	string	Yes	none
Content-Type	header	string	Yes	none
body	body	object	No	none
» contents	body	[object]	Yes	none
»» parts	body	[object]	No	none
»»» text	body	string	No	none

Response Example

200 Response

{
  "id": "string",
  "object": "string",
  "created": 0,
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "string",
        "content": "string"
      },
      "finish_reason": "string"
    }
  ],
  "usage": {
    "prompt_tokens": 0,
    "completion_tokens": 0,
    "total_tokens": 0
  }
}

Responses

Status	Meaning	Description	Model
200	OK	none	Inline

Response Schema

Status 200

Name	Type	Required	Constraints	Description
» id	string	true	none	none
» object	string	true	none	none
» created	integer	true	none	none
» choices	[object]	true	none	none
»» index	integer	false	none	none
»» message	object	false	none	none
»»» role	string	true	none	none
»»» content	string	true	none	none
»» finish_reason	string	false	none	none
» usage	object	true	none	none
»» prompt_tokens	integer	true	none	none
»» completion_tokens	integer	true	none	none
»» total_tokens	integer	true	none	none

Pham vi Apifox​

API model / Chat / OpenAI / Dinh dang Chat Completions​

POST Completions Legacy Format​

Request Parameters​

Details​

Responses​

Response Schema​

GET Token Usage Query​

Request Parameters​

Responses​

Response Schema​

API model / Chat / OpenAI / Dinh dang Responses​

POST xAI Web Search​

Request Parameters​

Responses​

Response Schema​

API model / Chat / Anthropic Claude / Dinh dang Claude goc (khuyen dung)​

POST Web Search​

Request Parameters​

Responses​

Response Schema​

API model / Chat / Google Gemini / Dinh dang Gemini goc​

POST Gemini Text Chat​

Request Parameters​

Responses​

Response Schema​

Pham vi Apifox

API model / Chat / OpenAI / Dinh dang Chat Completions

POST Completions Legacy Format

Request Parameters

Details

Responses

Response Schema

GET Token Usage Query

Request Parameters

Responses

Response Schema

API model / Chat / OpenAI / Dinh dang Responses

POST xAI Web Search

Request Parameters

Responses

Response Schema

API model / Chat / Anthropic Claude / Dinh dang Claude goc (khuyen dung)

POST Web Search

Request Parameters

Responses

Response Schema

API model / Chat / Google Gemini / Dinh dang Gemini goc

POST Gemini Text Chat

Request Parameters

Responses

Response Schema