Flink AI Native Ops will be officially commercialized in phases by region starting from October 13, 2026, with full-network coverage completed on October 26, 2026. This announcement describes service activation and billing, and the impact on existing resources.
Service changes
Flink AI Native Ops will be officially commercialized starting from October 13, 2026, rolled out in phases by region, with full-network coverage completed on October 26, 2026. The free trial benefits remain available. After the free trial quota is exhausted, AI Service must be activated to continue usage, and usage beyond the free quota is billed based on actual consumption.
Effective schedule: Overseas regions take effect from October 13, 2026 (first batch: London), and the remaining overseas regions are covered in batches on October 15 and October 19. Chinese mainland regions take effect from October 21 (excluding Shanghai and Hangzhou), and the remaining regions such as Shanghai and Hangzhou are covered on October 26. Before the effective date in your region, the related capabilities remain free under the current model.
Service activation and billing
-
Free trial: Each workspace is granted a one-time free trial quota of 30 AI Assistant uses and 5 Intelligent Inspection uses (not refreshed), shared across all projects in the workspace. All capabilities can be tried without activation. After the trial quota is exhausted, activate AI Service to continue.
-
Activate AI Service: AI Service is the unified managed model invocation service of Realtime Compute for Apache Flink. Model calls from capabilities such as AI Native Ops and SQL AI functions are metered and billed through AI Service. For service details, see Flink AI service (built-in models). After activation, each Alibaba Cloud account is granted a free quota of 100,000 tokens per month in each region, refreshed monthly. This quota is independent of and does not consume the free quota of AI Function and Flink Agents (1 million tokens per account per region per month). Usage beyond the free quota is billed by actual token consumption, with input and output priced separately. Multiple model series such as Qwen, DeepSeek, and GLM are supported, billed at tiered unit prices, with discounts applied on context cache hits. For billing details, see AI Service billing (pay-as-you-go).
-
Deactivate the service: Deactivation is supported at the feature level and the service level, and billing stops immediately after deactivation. At the feature level, scheduled inspection plans can be paused or stopped at any time. At the service level, AI Service can be deactivated. The service is ready to use upon activation, enabled on demand, and cost-controllable.
Impact on existing resources
-
Existing scheduled inspection plans, conversation records, and inspection reports are all retained. Enabled automatic inspection is set to disabled by default. A confirmation dialog stating the billing rules appears before an inspection plan is enabled, and no quota is deducted before confirmation.
-
After the free quota is exhausted, inspection plans are automatically paused with configurations retained, and automatically resume after AI Service is activated.
Additional notes
Continued use of Flink AI Native Ops after the billing rules take effect is deemed as acknowledgment of and agreement to the billing rules.