Activation and billing overview-AI Guardrails(AI Guardrails)-阿里云帮助中心

When using large language model (LLM) APIs, you must check user inputs and model outputs for content safety. This topic describes how to activate the pay-as-you-go service for Guardrails and details the billing rules for each moderation type.

Important

After you activate pay-as-you-go for Guardrails, the Content Compliance feature is enabled by default. To use the Sensitive Content Detection or prompt injection detection features, you must first enable them in the console, as described later in this topic.

Prerequisites

Before you can activate Guardrails, your Alibaba Cloud account must complete identity verification.

Activate Guardrails pay-as-you-go

Activating the pay-as-you-go service for Guardrails is free. The system bills you based on your actual usage. For specific unit prices, see Moderation types and unit prices.

Choose a suitable access method based on your business scenario:

API access: Call the moderation service directly using APIs. This method is ideal for using Guardrails as a standalone service.
AI Gateway integration: Manage moderation policies for multiple large models centrally through AI Gateway. This method is ideal for scenarios involving mixed-model calls.
Model Studio integration: Call Guardrails directly from the Model Studio platform. This method is ideal for existing Model Studio users.

Enable detection features

Log on to the Guardrails console. In the left-side navigation pane, choose Protection Config > Model Protection.
In the list of detection configurations, find the target service and click Configuration in the Actions column to open the service management page.
In the Protection Dimension section, use the toggle on each card to enable or disable the feature:
- Content Compliance: Enabled by default. Detects pornographic, violent, political, and other undesirable content. This feature is billed separately. For pricing details, see Moderation types and unit prices.
- Sensitive content detection: Detects potential leaks of personal information or sensitive corporate data. This feature is billed separately. For pricing details, see Moderation types and unit prices.
- Prompt injection detection: Detects malicious prompts designed to bypass the safety limits of large models. This feature is billed separately. For pricing details, see Moderation types and unit prices.
- Malicious URL (in public preview): Scans content from large models for malicious links. This feature is currently free to use during the public preview.
- Model Hallucination (in public preview): Detects false or inaccurate information generated by large models. This feature is currently free to use during the public preview.
- Detection agent: Identifies specific, user-defined risks in large model interactions based on flexible custom configurations. For pricing details, see detection agent billing plans.

Note

After a feature is enabled, its toggle appears in the enabled state.

Moderation types and unit prices

Access method	Type	Services	Unit price
API access/AI Gateway integration	Text Moderation - Per Call - Advanced text_guard_advanced	AI input content security check (Pro) (query_security_check_pro) AI-generated content security check (Pro) (response_security_check_pro) Important The query_security_check_pro and response_security_check_pro services are now generally available. They provide more granular labels for content compliance protection, with a significantly larger number of labels than the previous version. AI input content security check (query_security_check) AI-generated content security check (response_security_check) AI input content security check (International) (query_security_check_cb) AI-generated content security check (International) (response_security_check_cb)	CNY 15 per 10,000 calls
API access	Image Moderation - Per Call - Advanced image_guard_advanced	AIGC input image security check (img_query_security_check) AIGC output image security check (img_response_security_check)	CNY 30 per 10,000 calls
API access	Synchronous File Moderation - Per Call - Advanced file_guard_sync_advanced	Real-time file check (file_security_sync_check) Real-time multimodal (text file) check (text_file_sec_sync_check)	CNY 20 per 10,000 segments Important The text content in a file is divided into 2,000-character segments. You are billed based on the number of segments.
API access	Text-Image Moderation - Per Call - Advanced text_image_guard_advanced	multimodal (text and image) content security check (text_img_security_check)	CNY 45 per 10,000 calls
API access/AI Gateway integration	Sensitive Content Moderation - Per Call - Advanced text_sddp_advanced	AI input content security check (query_security_check) AI-generated content security check (response_security_check) AI input content security check (International) (query_security_check_cb) AI-generated content security check (International) (response_security_check_cb)	CNY 15 per 10,000 calls
API access/AI Gateway integration	prompt injection moderation text_guard_prompt_attack	AI input content security check (query_security_check) AI-generated content security check (response_security_check) AI input content security check (International) (query_security_check_cb) AI-generated content security check (International) (response_security_check_cb)	CNY 15 per 10,000 calls Important This feature will be commercially billed starting January 22, 2026. If you do not require this service, disable it in the Guardrails console to avoid incurring charges.
API access	Malicious File Detection	Real-time file check (file_security_sync_check)	This feature is in public preview. After you enable it, you can use it for free.
API access	digital watermarking	AIGC output image security check (img_response_security_check)	This feature is in public preview. After you enable it, you can use it for free.
Model Studio integration	Text Compliance Moderation - Per Token - Advanced text_guard_token_advanced	Model Studio input content security guardrail_Qwen3Guard_pro (qwen_query_check_pro) Model Studio output content security guardrail_Qwen3Guard_pro (qwen_response_check_pro) Model Studio input content security guardrail_pro (bl_query_guard_pro) Model Studio output content security guardrail_pro (bl_response_guard_pro)	CNY 0.003 per 1,000 tokens
Model Studio integration	Text Compliance Moderation - Per Token - Standard text_guard_token_standard	Model Studio input content security guardrail_Qwen3Guard (qwen_query_check) Model Studio output content security guardrail_Qwen3Guard (qwen_response_check) Model Studio input content security guardrail (bl_query_guard) Model Studio output content security guardrail (bl_response_guard)	CNY 0.0004 per 1,000 tokens
Model Studio integration	Sensitive Content Moderation - Per Token - Advanced text_sddp_token_advanced	Model Studio input content security guardrail_Qwen3Guard_pro (qwen_query_check_pro) Model Studio output content safety guardrail: Qwen3Guard Pro (`qwen_response_check_pro`) Model Studio input content security guardrail_pro (bl_query_guard_pro) Model Studio output content security guardrail_pro (bl_response_guard_pro)	CNY 0.003 per 1,000 tokens
Model Studio integration	Sensitive Content Moderation - Per Token - Standard text_sddp_token_standard	Model Studio Input Content Safety Guardrail_Qwen3Guard Edition (qwen_query_check) Model Studio Output Content Safety Guardrail_Qwen3Guard Edition (qwen_response_check) Model Studio Input Security Guardrail (bl_query_guard) Model Studio output content security guardrail (bl_response_guard)	CNY 0.0004 per 1,000 tokens
Model Studio integration	prompt injection moderation	Model Studio input content security guardrail_Qwen3Guard_pro (qwen_query_check_pro) Model Studio output content security guardrail_Qwen3Guard_pro (qwen_response_check_pro) Model Studio input content security guardrail_Qwen3Guard (qwen_query_check) Model Studio output content security guardrail_Qwen3Guard (qwen_response_check) Model Studio input content security guardrail_pro (bl_query_guard_pro) Model Studio output content security guardrail_pro (bl_response_guard_pro) Model Studio input content security guardrail (bl_query_guard) Model Studio output content security guardrail (bl_response_guard)	CNY 15 per 10,000 calls Important This feature will be commercially billed starting January 22, 2026. If you do not require this service, disable it in the Guardrails console to avoid incurring charges.
Model Studio integration	Image Moderation - Per Call - Advanced image_guard_advanced	Model Studio input image security guardrail (bl_img_query_guard) Model Studio output content security guardrail (bl_img_response_guard)	CNY 30 per 10,000 calls

Important

When you use Guardrails through Model Studio integration for a single query or response check, if a text contains fewer than 1,000 tokens, it is billed as 1,000 tokens. If the text contains 1,000 or more tokens, you are billed for the actual number of tokens.

With pay-as-you-go, Guardrails bills you based on the type and volume of content you check.

Go to the Guardrails activation page.
Read and agree to the Terms of Service, then click Buy Now.
After activation, log on to the Guardrails console and confirm that the service status is Activated.

How can I confirm that pay-as-you-go has been activated for my Guardrails account?

You can use either of the following methods to verify the activation status:

Log on to the Guardrails console and check whether the service status shows Activated.
Log on to the Alibaba Cloud Management Console and go to Expenses and Costs > Order Management. Search for "AI Guardrails" and check whether an activation record exists.

Note

Guardrails uses the pay-as-you-go billing method. Activating the service is free of charge. You are only billed when you call the API operations, so you do not need to worry about unexpected charges after activation.

Query resource plan usage and remaining quota: in the Alibaba Cloud User Center, go to the Resource Instance Management page to view the details of your Content Moderation resource plan, its resource plan usage, and the remaining number of calls. You can also open the Expenses and Costs console to view usage and bill details. If no data is returned, verify that the current account has purchased a Content Moderation resource plan, and check the purchase records under that account.
Troubleshoot region and account mismatches: if the regions available when you activate the service in the console differ from the regions available for API calls and the call fails, check whether the account that you use for the API call is the same as the account that activated Guardrails. If the accounts do not match, switch to the correct account that has the service activated and call the operation again, or activate the service again under the account that returns the error.

Overdue payments and top-ups

The billing cycle window for Guardrails is 5 minutes.

After each billing cycle, Alibaba Cloud generates a bill based on your usage from the previous cycle and automatically deducts the amount from your account balance.

Billing account and related services

Guardrails fees are deducted directly from the balance of your Alibaba Cloud account. Guardrails and Model Studio share the balance of the same Alibaba Cloud account, so the funds are interchangeable between the two services. No separate Model Studio balance exists.

Activating Guardrails does not automatically activate monitoring services such as Application Real-Time Monitoring Service (ARMS) or OpenTelemetry. You must activate each of those services separately. If you do not activate them, they generate no charges.

If your account has an overdue payment, your service is suspended. The service resumes within five minutes after you pay the outstanding balance. For instructions on how to add funds to your account, see the Top-up guide.

FAQ

How do I cancel an order, unsubscribe, release resources, or disable the service to stop billing?

Guardrails pay-as-you-go does not support directly disabling the service, unsubscribing, or canceling an order that is already activated. Activation itself is free of charge. To stop billing, stop calling all detection API operations in your business code. After you stop all detection API calls, no new charges are generated.