Flink AI service (built-in models) - Realtime Compute for Apache Flink

Overview

Flink AI service (built-in models) is a managed AI model invocation capability of Realtime Compute for Apache Flink. You can call built-in models over the private network in Flink jobs without configuring an API-Key or PrivateLink, enabling streaming AI inference and embedding.

Core advantages

Feature	Description
Out-of-the-box	No need to configure an API-Key, endpoint, or PrivateLink.
Billed by token	Token-based billing with clear, controllable costs.
Multi-model support	Multiple mainstream large models are built in, covering scenarios such as text generation, visual understanding, translation, and embedding.
Global region coverage	Available in multiple regions inside and outside the Chinese mainland, with automatic cross-region access.

How it works

Enable and disable the service

Enable the service

Log on to the management console of Realtime Compute for Apache Flink. On the Flink AI Service page, click Activate.

The service must be enabled by a primary account or a RAM user that is granted the AliyunBSSOrderAccess and AliyunStreamFullAccess permissions.
Once enabled in a region, all workspaces in that region can use it.
This feature is pay-as-you-go. You are charged only for model invocations, not for enabling the feature. For details, see AI service billing (pay-as-you-go).

Disable the service

On the Flink AI Service page, click Disable Service and confirm the following:

After you disable the feature, all jobs in the region can no longer use the built-in model service.
Before you disable the feature, make sure that no jobs depend on the built-in models.

Service protection

To prevent unexpected business interruptions caused by accidental disabling of the service, you can enable service protection in the console.

After service protection is enabled, the system rejects all manual release requests initiated through the console, API, or CLI. This feature is supported only in the pay-as-you-go mode of Flink AI service and applies only to manual release operations. It does not take effect in the following scenarios in which the system automatically releases the service:

The service is disabled due to overdue payments on the account.
The service is forcibly released due to violations of cloud platform security rules.

Supported regions

International site

Flink AI service region	Region ID	Endpoint region	Inference execution region
China (Beijing)	cn-beijing	China (Beijing)	Chinese mainland (dynamic scheduling)
China (Zhangjiakou)	cn-zhangjiakou
China (Shanghai)	cn-shanghai
China (Hangzhou)	cn-hangzhou
China (Shenzhen)	cn-shenzhen
China (Chengdu)	cn-chengdu
Singapore	ap-southeast-1	Singapore	Regions outside the Chinese mainland (dynamic scheduling)
China (Hong Kong)	cn-hongkong
Malaysia (Kuala Lumpur)	ap-southeast-3
Indonesia (Jakarta)	ap-southeast-5
Japan (Tokyo)	ap-northeast-1
Korea (Seoul)	ap-northeast-2
Thailand (Bangkok)	ap-southeast-7
UAE (Dubai)	me-east-1
Mexico	na-south-1
US (Virginia)	us-east-1
US (Silicon Valley)	us-west-1

Important

The region determines the physical location of data access and inference execution. Select a region based on your data compliance requirements.

Endpoint region: Determines the access point and data storage location.
Inference execution region: Where inference runs. Dynamically scheduled within a defined scope.

When you select a region outside the Chinese mainland, your business data is transferred across borders to the Singapore region and, as needed, dynamically scheduled to inference service nodes outside the Chinese mainland for computation. You must ensure that such cross-border data transfers comply with all applicable laws, regulations, and regulatory requirements, including but not limited to: the data being transferred does not contain any content that applicable laws restrict or prohibit from cross-border transfer; all necessary approvals or filing procedures have been completed; and all required notification obligations have been fulfilled and necessary consents have been obtained.

Built-in models

Model family	Model	Capabilities	Supported regions
Qwen	qwen3.8-max	Text generation, reasoning, visual understanding	All
	qwen3.7-max	Text generation, reasoning, visual understanding	All
	qwen3.7-plus	Text generation, reasoning, visual understanding	All
	qwen3.6-plus	Text generation, reasoning, visual understanding	All
	qwen3.5-plus	Text generation, reasoning, visual understanding	All
	qwen3.7-flash	Text generation, reasoning, visual understanding	All
	qwen3.6-flash	Text generation, reasoning, visual understanding	All
	qwen3.5-flash	Text generation, reasoning, visual understanding	All
	qwen3.5-ocr	OCR	Chinese mainland
	qwen-vl-ocr	OCR	All
	qwen-mt-plus	Translation	All
	qwen-mt-flash	Translation	All
DeepSeek	deepseek-v4-flash-0731	Text generation, reasoning	All
DeepSeek	deepseek-v4-pro	Text generation, reasoning	All
GLM	glm-5.2	Text generation, reasoning	All
Embedding	qwen3.7-text-embedding	Text embedding	Chinese mainland
	text-embedding-v4	Text embedding	All
	qwen3-vl-embedding	Multimodal embedding	Chinese mainland

Usage

SQL

Specify the task and model parameters in the CREATE MODEL statement. You do not need to specify endpoint or api-key.

CREATE MODEL model_name
INPUT (column_name STRING)
OUTPUT (column_name {STRING | ARRAY<FLOAT>})
WITH (
  'provider' = 'openai-compat | dashscope',
  'task' = 'chat/completions | embeddings | multimodal-embedding',
  'model' = '<model-name>'
);

For more details, see Model configuration.

Pyflink Dataframe

Specify task in set_model_provider and model in the function. Example:

pf.set_model_provider(pf.OpenAICompatProvider(task="chat/completions"))

questions = pf.from_dict({
    "id": [1, 2],
    "question": ["What is Flink?", "What is stream processing?"],
})

# default output column: output (STRING)
df = questions.llm.predict("question", model="qwen3.6-plus")

For more information, see AI/LLM functions.