⌘K

GLM-5.3-Flash

glm-5.3-flash

GLM-5.3-Flash is Z.AI's first native multimodal model in the GLM-5 family. It accepts video, image, text, and file inputs, supports a 1M-token context window and up to 131,072 output tokens, and always enables thinking through an OpenAI Chat Completions-compatible API.

Authentication

authorization string required

All APIs require authentication via Bearer Token.

Get API Key:

Visit API Key Management Page to get your API Key.

Usage:

Add to request header:

Authorization: Bearer YOUR_API_KEY

Endpoint and compatibility

POST https://cloud.vtrix.ai/llm/chat/completions

This endpoint uses the OpenAI Chat Completions format: the top level contains choices and usage. It is not an asynchronous task API, so the response does not contain status, output, asset url, jobId, cost, or discount fields.

Model capabilities

GLM-5.3-Flash is a native multimodal model that accepts video, image, text, and file inputs, with a 1M-token context window and a maximum output of 131,072 tokens. For multimodal requests, use content blocks such as image_url, video_url, and file. Thinking is always enabled: thinking.type must be enabled, and reasoning_effort supports low, high, and max (default: max).

Parameters

model string required

Model ID for the request.

Value: glm-5.3-flash

messages array required

Conversation history.

role string required

One of user, assistant, system, or tool.

content string | array required

A text string or an OpenAI-compatible content array containing text, image, video, or file items. Use image_url, video_url, or file content blocks for multimodal inputs.

thinking object

GLM-5.3-Flash always enables thinking. thinking.type must be enabled; disabling thinking is not supported. For preserved thinking across turns, set clear_thinking to false.

type string

enabled.

clear_thinking boolean

Whether to preserve reasoning content from previous turns.

Default: true. Set to false to preserve thinking blocks across turns.

reasoning_effort string

Reasoning effort for always-on thinking. Default: max.

Options: low, high, max.

max_tokens integer

Maximum completion token budget, from 1 to 131072. Thinking tokens consume this budget, so reserve sufficient tokens when thinking is enabled.

temperature number

Sampling temperature. Default: 1.0; range: 0.0 to 1.0.

top_p number

Nucleus sampling parameter. Default: 0.95; range: 0.01 to 1.0.

stream boolean

Streams Server-Sent Events (SSE) when true; default: false. Stream chunks use the OpenAI Chat Completions chunk format and end with data: [DONE].

tool_stream boolean

Whether to stream function-call arguments incrementally.

Default: false.

tools array

Tools available to the model, using the OpenAI function tool structure.

type string required

function.

function object required

Function definition.

tool_choice string | object

Tool selection strategy: auto, none, required, or an explicit function selection.

stop string | array

One or more stop sequences; up to four strings.

response_format object

Output-format constraint. Its type may be text or json_object.

Response format

id string

Chat Completion request ID.

object string

Always chat.completion.

created integer

Unix timestamp in seconds for response creation.

model string

Model ID that processed the request.

choices array

One or more completion choices.

index integer

Choice index.

message object

Assistant message.

role string

Always assistant.

content string | null

Final answer text.

reasoning_content string (optional)

Reasoning text produced by GLM-5.3-Flash’s always-on thinking.

finish_reason string

Completion reason, such as stop, length, or tool_calls.

usage object

Token usage for the request.

prompt_tokens integer

Input tokens.

completion_tokens integer

Output tokens.

total_tokens integer

Sum of input and output tokens.

completion_tokens_details.reasoning_tokens integer (optional)

Tokens consumed by thinking.