Chat Completions

Authorizations

Authorization

string

header

required

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

Body

application/json

model

string

required

The model ID used to generate the reply, such as gpt-5.6-sol or o3.

messages

object[]

required

List of conversation messages as of current date.

Minimum array length: 1

Hide child attributes

messages.role

enum<string>

required

Message author role. Newer inference models recommend using developer instead of system.

Available options:

developer,

system,

user,

assistant,

tool,

function

messages.name

string

Optional participant name used to distinguish different participants under the same role.

messages.content

Message content. It can be a string, or an array composed of text, pictures, audio, and file content blocks. The assistant tool call message can be null.

messages.tool_call_id

string

When role is tool, the tool call ID to which this message responds.

messages.tool_calls

object[]

List of tool calls generated by the model when role is assistant.

Hide child attributes

messages.tool_calls.id

string

Tool call ID.

messages.tool_calls.type

enum<string>

Tool type.

Available options:

function,

custom

messages.tool_calls.function

object

Hide child attributes

messages.tool_calls.function.name

string

function name.

messages.tool_calls.function.arguments

string

Function parameters as JSON strings.

messages.tool_calls.custom

object

Hide child attributes

messages.tool_calls.custom.name

string

Custom tool name.

messages.tool_calls.custom.input

string

Custom tool input.

messages.function_call

object

Legacy function call information, deprecated, please use tool_calls.

Hide child attributes

messages.function_call.name

string

function name.

messages.function_call.arguments

string

Function parameters as JSON strings.

messages.refusal

string

Assistant's rejection message.

messages.audio

object

The message of the previous model audio reply.

Hide child attributes

messages.audio.id

string

Audio response ID.

audio

object | null

Audio output parameters. Required when modalities contains audio.

Hide child attributes

audio.format

enum<string>

Output audio format.

Available options:

wav,

aac,

mp3,

flac,

opus,

pcm16

audio.voice

The sound used by the model can be a built-in sound name or a custom sound object.

temperature

number | null

Sampling temperature. Higher values make the output more random; usually adjusted with top_p alternatively.

Required range: 0 <= x <= 2

top_p

number | null

Kernel sampling threshold; usually adjusted alternatively with temperature.

Required range: 0 <= x <= 1

integer | null

The number of candidate responses generated for each input.

Required range: x >= 1

stream

boolean | null

Whether to return incremental results in SSE streaming.

stream_options

object | null

Additional configuration for streaming output, only takes effect when stream is true.

Hide child attributes

stream_options.include_usage

boolean

Whether to return usage information in the final statistics block.

max_completion_tokens

integer | null

The maximum number of generated tokens, including visible output and inference tokens.

Required range: x >= 1

max_tokens

integer | null

Deprecated, use max_completion_tokens instead.

Required range: x >= 1

frequency_penalty

number | null

Frequency penalty. Positive values reduce duplicate expressions.

Required range: -2 <= x <= 2

presence_penalty

number | null

There are penalties. This is a great time to encourage models to introduce new content.

Required range: -2 <= x <= 2

logit_bias

object | null

Specifies the generation tendency of the word element. The attribute is named token ID and has values from -100 to 100.

Hide child attributes

logit_bias.{key}

number

Required range: -100 <= x <= 100

logprobs

boolean | null

Whether to return the log probability of the output token.

top_logprobs

integer | null

The number of high-probability candidate tokens returned for each output position; logprobs needs to be enabled at the same time.

Required range: 0 <= x <= 20

stop

Stop sequence, some new models do not support it.

response_format

object

Output format configuration.

Option 1
Option 2
Option 3

Hide child attributes

response_format.type

enum<string>

required

Available options:

text

tools

object[]

A list of callable tools provided to the model.

Function tools.

Option 1
Option 2

Hide child attributes

tools.type

enum<string>

required

Available options:

function

tools.function

object

required

Hide child attributes

tools.function.name

string

required

Maximum string length: 64

tools.function.description

string

tools.function.parameters

object

JSON Schema of function parameters.

tools.function.strict

boolean | null

tool_choice

Controls whether the model calls the tool, automatically selects the tool, or forces the specified tool to be called.

Available options:

none,

auto,

required

parallel_tool_calls

boolean

Whether the model is allowed to call multiple tools in parallel.

reasoning_effort

enum<string> | null

Inference strength settings for trade-offs between speed, cost, and inference depth.

Available options:

none,

minimal,

low,

medium,

high,

xhigh,

max

verbosity

enum<string> | null

Output verbosity.

Available options:

low,

medium,

high

seed

integer | null

A random seed used to try to make the output reproducible.

service_tier

enum<string> | null

The requested service level.

Available options:

auto,

default,

flex,

scale,

priority,

fast

metadata

object | null

Additional key-value pair metadata.

Hide child attributes

metadata.{key}

string

modalities

enum<string>[] | null

Specify the output type.

Available options:

text,

audio

prediction

object | null

Prediction output configuration.

Hide child attributes

prediction.type

enum<string>

required

Available options:

content

prediction.content

required

Expected output content.

prompt_cache_key

string | null

Stable key for hint word cache.

prompt_cache_retention

enum<string> | null

The length of time the prompt word cache is retained.

Available options:

in_memory,

24h

prompt_cache_options

object

Prompt word caching options.

Hide child attributes

prompt_cache_options.mode

enum<string>

cache mode.

Available options:

implicit,

explicit

prompt_cache_options.ttl

enum<string>

Cache validity period.

Available options:

30m

safety_identifier

string | null

A stable end-user identifier used to assist in detecting abuse and should not be directly personally identifiable information.

user

string

Old version of end user identification, it is recommended to use safety_identifier.

web_search_options

object

Web search tool configuration.

Hide child attributes

web_search_options.search_context_size

string

Search context size.

web_search_options.user_location

object

Approximate location of the user, used to improve localized search relevancy.

Hide child attributes

web_search_options.user_location.type

enum<string>

Available options:

approximate

web_search_options.user_location.country

string

Country code.

web_search_options.user_location.city

string

City name.

web_search_options.user_location.region

string

Region or province/state name.

web_search_options.user_location.timezone

string

Time zone identifier.

function_call

deprecated

Deprecated, use tool_choice.

Available options:

none,

auto

functions

object[]

deprecated

Deprecated, please use tools.

Hide child attributes

functions.name

string

required

Maximum string length: 64

functions.description

string

functions.parameters

object

JSON Schema of function parameters.

moderation

object | null

Configuration to perform auditing on request input and build output.

Hide child attributes

moderation.model

string

Audit model ID, such as omni-moderation-latest.

moderation.policy

object | null

Input and output audit policies.

Hide child attributes

moderation.policy.input

object | null

Hide child attributes

moderation.policy.input.mode

enum<string>

Available options:

score,

block

moderation.policy.output

object | null

Hide child attributes

moderation.policy.output.mode

enum<string>

Available options:

score,

block

enable_thinking

boolean

Qwen/Bailian extension: enable reasoning when supported by the model.

thinking_budget

integer

Qwen/Bailian extension: reasoning token budget. Some models cannot combine this with reasoning_effort.

Required range: x >= 0

thinking

object

Reasoning configuration for GLM, Bailian, and MiniMax models. Supported modes vary by model.

Hide child attributes

thinking.type

enum<string>

required

Available options:

enabled,

disabled,

adaptive

thinking.clear_thinking

boolean

preserve_thinking

boolean

Bailian extension: retain earlier reasoning content on supported models.

clear_thinking

boolean

Bailian extension: clear earlier reasoning content.

top_k

integer | null

Qwen/Bailian extension: sampling candidate count.

Required range: x >= 0

repetition_penalty

number

Qwen/Bailian extension: repetition penalty; 1 disables the penalty.

Required range: x > 0

do_sample

boolean

GLM extension: enable sampling. When false, temperature and top_p are ignored.

tool_stream

boolean

Qwen/GLM extension: stream tool-call arguments with stream=true.

enable_search

boolean

Qwen/Bailian extension: allow web search.

search_options

object

Qwen/Bailian search settings, used when enable_search=true.

Hide child attributes

search_options.forced_search

boolean

search_options.search_strategy

enum<string>

Available options:

turbo,

max,

agent

search_options.enable_source

boolean

search_options.enable_citation

boolean

search_options.citation_format

string

search_options.enable_search_extension

boolean

enable_code_interpreter

boolean

Qwen/Bailian extension: enable the code interpreter where available.

vl_high_resolution_images

boolean

Qwen VL extension: enable high-resolution image understanding.

skill

object[]

Bailian qwen-doc-turbo generation skills; stream must be true.

Minimum array length: 1

Hide child attributes

skill.type

string

required

Allowed value: "ppt"

skill.mode

enum<string>

Available options:

general,

creative

skill.template_id

enum<string>

Available options:

news_01,

summary_01,

internet_01,

thesis_01

Response

200 - application/json

Successful response

Chat Completions non-streaming responses (chat.completion). "Generate via JSON etc. → JSON Schema" direct import for Apifox.

string

required

The unique ID of this chat completion.

object

enum<string>

required

Object type, fixed to chat.completion.

Available options:

chat.completion

created

integer

required

Unix timestamp of the time the response was created, in seconds.

model

string

required

The actual model ID used to generate the reply.

choices

object[]

required

A list of candidate responses generated by the model.

Hide child attributes

choices.index

integer

required

The sequence number of the candidate reply, starting from 0.

choices.finish_reason

enum<string> | null

required

The reason for the end of the generation: stop is the natural end, length is the length limit reached, tool_calls is the tool call, and content_filter is content filtering.

Available options:

stop,

length,

tool_calls,

content_filter,

function_call,

null

choices.message

object

required

Hide child attributes

choices.message.role

enum<string>

required

Message author role, fixed to assistant.

Available options:

assistant

choices.message.content

string | null

Text content generated by the model; may be null when the tool is called.

choices.message.refusal

string | null

Description generated when the model rejects a request.

choices.message.annotations

object[]

Reply to comments such as URL references.

Hide child attributes

choices.message.annotations.type

string

Annotation type, such as url_citation.

choices.message.annotations.url_citation

object

Hide child attributes

choices.message.annotations.url_citation.start_index

integer

The starting character index of the quoted text.

choices.message.annotations.url_citation.end_index

integer

The ending character index of the quoted text.

choices.message.annotations.url_citation.title

string

Quote the page title.

choices.message.annotations.url_citation.url

string

The URL of the referring page.

choices.message.audio

object | null

Audio output related data.

Hide child attributes

choices.message.audio.id

string

The unique ID of the audio response.

choices.message.audio.data

string

Base64 encoded audio data.

choices.message.audio.expires_at

integer

Unix timestamp of the audio data expiration time, in seconds.

choices.message.audio.transcript

string

Text transcription of audio content.

choices.message.function_call

object | null

Legacy function call information, deprecated.

Hide child attributes

choices.message.function_call.name

string

function name.

choices.message.function_call.arguments

string

Function parameters as JSON strings.

choices.message.tool_calls

object[]

A list of tools called by the model request.

Hide child attributes

choices.message.tool_calls.id

string

The unique ID of the tool call.

choices.message.tool_calls.type

enum<string>

Tool type.

Available options:

function

choices.message.tool_calls.function

object

Hide child attributes

choices.message.tool_calls.function.name

string

The name of the function called by the model.

choices.message.tool_calls.function.arguments

string

Function parameters in the form of JSON strings should be parsed and verified before use.

choices.logprobs

object | null

Outputs log-probability information for tokens; only returned if logprobs is enabled on request.

Hide child attributes

choices.logprobs.content

object[] | null

Output the probability information of each token in the content.

Hide child attributes

choices.logprobs.content.token

string

Tome text.

choices.logprobs.content.bytes

integer[] | null

The UTF-8 byte sequence corresponding to the token.

choices.logprobs.content.logprob

number

The logarithmic probability of this token.

choices.logprobs.content.top_logprobs

object[]

The most likely candidate token for this position.

Hide child attributes

choices.logprobs.content.top_logprobs.token

string

choices.logprobs.content.top_logprobs.bytes

integer[] | null

choices.logprobs.content.top_logprobs.logprob

number

usage

object

Word usage statistics for this request.

Hide child attributes

usage.prompt_tokens

integer

Enter the number of tokens used in the prompt word.

usage.completion_tokens

integer

The number of tokens used by the model to generate content.

usage.total_tokens

integer

The total number of tokens requested.

usage.prompt_tokens_details

object

Enter the word element details.

Hide child attributes

usage.prompt_tokens_details.cached_tokens

integer

The number of tokens in the input that hit the cache.

usage.prompt_tokens_details.audio_tokens

integer

The number of audio tokens in the input.

usage.prompt_tokens_details.image_tokens

integer

Number of image tokens in the input.

usage.prompt_tokens_details.text_tokens

integer

The number of text tokens in the input.

usage.prompt_tokens_details.cache_write_tokens

integer

The number of original tokens written to the prompt word cache.

usage.completion_tokens_details

object

Generate token details.

Hide child attributes

usage.completion_tokens_details.reasoning_tokens

integer

The number of tokens used by the model for inference.

usage.completion_tokens_details.audio_tokens

integer

The number of audio tokens generated by the model.

usage.completion_tokens_details.text_tokens

integer

The number of text tokens generated by the model.

usage.completion_tokens_details.accepted_prediction_tokens

integer

The number of predicted tokens that actually appear in the generated results.

usage.completion_tokens_details.rejected_prediction_tokens

integer

The number of predicted tokens that did not appear in the generated results.

service_tier

string | null

The actual service level used to handle this request.

system_fingerprint

string | null

The backend configuration fingerprint for the model to run on.

metadata

object

Metadata associated with the request.

Hide child attributes

metadata.{key}

string