Billing
Billing items, free items, and free quotas for Flow Agent agent applications and workflow applications.
Flow Agent does not charge for orchestration itself. Costs come from the models and components your application actually calls. Each component is billed independently at its own published rate, free quotas only offset their specified billing items, and the actual amount is shown on your Alibaba Cloud bill.
Billing items
Costs originate differently for each application type.
| Application type | Cost sources | Details |
|---|---|---|
| Agent application (Agent 2.0) | Model calls, knowledge base, MCP calls | Agent application billing |
| Workflow application | Call volume of the billable nodes on the execution path | Workflow application billing |
Agent application (Agent 2.0)
- Model calls: Billed by the input and output tokens of the selected model. For rates, see Recommended models. In Smart Mode, AUTO is billed at a unified list price regardless of which model the request is routed to; when you select the Performance, Balanced, or Economy tier directly, billing follows that tier. For tier rates and cache rates, see Smart Mode pricing details.
- Knowledge base: Billed on a pay-as-you-go basis. See Knowledge base pricing. Text chunks retrieved from the knowledge base increase model input tokens, which may raise model call costs.
- MCP: Some official MCP services are billed based on model invocations, such as text-to-image, text-to-video, and speech synthesis. Some MCP services involve third-party API calls; those fees are charged by the third party, and Alibaba Cloud Model Studio does not charge any additional fees.
- Context cache: Implicit caching takes effect automatically when the selected model supports it, requires no configuration, and cannot be disabled. Agent applications do not yet support configuring explicit cache. Smart Mode cache charges follow the corresponding tier prices in Smart Mode pricing details. For directly selected models, see Context cache for model eligibility and cache billing rules.
Workflow application
A workflow application is billed by the call volume of the billable nodes it actually executes.
- Billing unit: Each billable node has its own billing unit. For example, a large model node is billed by tokens.
- Total cost: The total cost of a single workflow call is the sum of the costs of all billable nodes on that execution path. Nodes that are not on the execution path incur no cost.
Note the billing behavior of the following nodes when you configure them.
| Node | Billing |
|---|---|
| Image parsing | The LLM document parser is currently free; the Qwen-VL parser incurs charges for Qwen-VL model inference. See Image parsing node |
| Document, audio, and video parsing | Parsed by built-in large models; currently free |
| Model built-in web search | Incurs model call fees and search strategy fees. See Web search |
| Knowledge base | Billed at the pay-as-you-go rate of the knowledge base. See Knowledge base pricing |
Pro-code applications are outside the two Flow Agent application types. For deployment and resource configuration, see Pro-code applications and Sandbox, and check the billing rules of the corresponding services.
Free items
| Feature | Description |
|---|---|
| File upload | Uploading files in file Q&A is free of charge. Q&A based on file content is billed at the standard model rate. See File Q&A |
| Automatic prompt optimization | Free of charge. See Optimize prompts |
| UI designer | The feature itself is free. AI conversations incur model invocation fees; file storage and database usage beyond the free quota or plan quota is billed pay-as-you-go. Publishing to production requires a Team Edition or higher subscription plan. See UI designer billing |
Free quota
Model calls can be offset by the free quota for new users. After the quota is used up, you are billed by token usage. Free quotas for components such as knowledge base and ParseX follow their own rules and only offset their specified billing items. See the billing documentation of each component.
FAQ
Who pays for costs incurred through a shared link?
All costs incurred through a shared link are charged to the UID of the application creator. See Share an application.
How do I check the actual cost of an application?
See Bills and cost management. For a workflow application, reconcile the call volume of each billable node against the execution path.