Model usage
View usage for Alibaba Cloud Model Studio models.
For more information about free quotas, see New user free quota. Manage free quota usage on the Free Quota page.
Availability
- Regions and date range: Model usage data is available in all regions for September 2026 and later months. In the US (Virginia) region, call count data is available only for October 2026 and later months.
- Models: All models in the Model list, including fine-tuned models that are based on them.
- Scope: Token usage and similar metrics are used for cost verification and count only model calls billed on token pay-as-you-go. Usage for model training and calls through provisioned throughput units or dedicated throughput units is not included; call count includes provisioned and dedicated throughput unit calls. Tool usage generated by model calls is counted separately.
View free quota usage
The top of the Free Quota page provides the "Free Quota Usage Overview", which summarizes the free quota consumption of each model in the current workspace by dimensions such as total number of models, sufficient quota, usage above 50%, usage above 80%, and no free quota. It also displays a "Top 3 models with nearly exhausted free quota" list, helping you quickly identify models that require attention.
The "All Models" table lists each model's model Code, remaining free quota, expiration time, and status (not enabled/enabled/enablement not supported), as well as operations. It supports searching, sorting, and filtering to find specific models. Click a row to toggle the "Free Quota Only" switch or open the model details drawer.
Batch operation entries at the top of the page let you batch enable or batch disable "Free Quota Only" for the selected models, or enable/disable it for all models with one click. A confirmation prompt appears before the operation, success and failure results are displayed after the operation is complete, and you can exit batch operations afterwards. In the operation column, you can also purchase resource packages through Buy Savings Plans or Discounted Order. If your account has no valid payment method bound, batch operations will fail with a prompt to bind a bank card.
View free quota usage in the console
- Go to the Free Quota page and select a model type tab to view the free quota usage of each model.
- In the All Models table, find specific models by searching, sorting, and filtering. You can toggle the Free Quota Only switch or use batch operations to manage free quota settings.
NoteFree Quota Only: After you enable this feature, the service automatically stops when the free quota is exhausted (returning a 403 error: AllocationQuota.FreeTierOnly), preventing charges beyond the free quota. If you want to continue using model services after the free quota is exhausted and pay based on actual usage, keep this feature disabled. You can enable this feature only while your account still has unconsumed free quota. Once enabled, the feature can be disabled only after the free quota is fully consumed (free quota consumption records can be viewed in bills, which are aggregated and generated by minute; console free quota data is updated at minute-level and supports manual page refresh for synchronization. The free quota value displayed in the console prevails.)
View model usage
View usage on the Model Usage page. Data is aggregated by workspace, not by Alibaba Cloud account. For more information, see How do I check the total token usage for my Alibaba Cloud account?. Call count usually has lower latency than other usage metrics: call count can appear about 1 minute after a call, while token usage and costs may take about 10 minutes to update. Call count is not displayed for batch offline inference or batch online inference.
- Select a statistical period and time range. You can view data by minute, hour, day, or month: minute supports up to 1 day, hour up to 7 days, day up to 31 days, and month up to 12 months. When you view data by minute or hour, grouping is not supported in the trend chart.
- Select a grouping dimension: by model, API Key, workspace, region, service site, or specification. Specification includes factors that affect unit price, such as input/output/cache hits, context length, thinking mode, and video generation resolution.
- Filter by model, API Key, workspace, region, service site, call source, and inference type. Inference types include online inference, batch offline inference (Batch File), and batch online inference (Batch Chat). If the workspace has no batch usage data, only online inference is displayed.
- View the metric cards: Call Count, Tokens Usage, Image Count, Video/Audio Duration, Speech Synthesis Characters, Tool Call Count, and others. Cards are aggregated by billing unit. For example, text models and omni-modal models are both billed by tokens, so their usage is counted in Tokens Usage. Switch between Total and Average per Call. Average per Call is usage divided by successful calls.
- In the Model Usage Details table, view the model type, call count, model usage, and tool usage for each model. Click Details to view the summary and trend of different usage types and tool usage for the model. Click View Cost to view the payable amount for the usage.
- The Total Successful Calls section at the bottom lists all models in descending order of call count. You can view it in full screen and download the data.
Note
- Failure count, failure rate, throttling error count, and content safety interception count have moved to the Model Monitoring page and are no longer shown in model usage details.
- The previous Batch inference category is now split into batch offline inference (Batch File) and batch online inference (Batch Chat).
NoteOnly usage data for September 2026 and later months is available. For earlier data, go to Expenses and Costs.
View billing overview
On the Cost Overview page, view costs for model calls and model training, including total cost, subscription cost, bill cost, and bill trend. For details, see Billing and cost management.
Usage units
The metric cards on the Model Usage page are aggregated by billing unit. The usage units for different models are:
Type | Subcategory | Unit | Billing (invocation) |
|---|---|---|---|
Large language model | Billed by the number of tokens for input and output. | ||
Visual model | Images | Billed by the number of images successfully generated. | |
Video generation | Seconds | Billed by the number of video seconds successfully generated. | |
Speech model | Second, character, or token | May be billed by audio duration (seconds), corresponding text characters, or tokens, depending on the model. | |
Omni-modal model | Token | Text is billed by token count. Other modalities (audio, image, video) are billed by their corresponding token counts. | |
Embedding model | Token | Billed by the number of tokens in the input text. | |
Text embedding |
Going live
Recommendations for managing model usage:
- Control model output length: Limit the maximum length of content generated per invocation, and thus control costs, by reasonably limiting the thinking length and setting the
max_tokensparameter. - Select models based on task type: For simple jobs such as categorization and summary, prioritize lower-cost, lightweight models over powerful but more expensive models.
- Monitoring and alerting: Monitor usage trends. Set usage alerts to receive timely notifications for abnormal usage.
- Optimize prompts: Concise and clear prompts improve model output quality and reduce unnecessary input tokens.
- Use batch inference: For non-real-time, large-batch processing tasks, batch inference is often more cost-effective than real-time invocation.
Glossary
Noun | Explanation |
|---|---|
Token | A token is the basic unit that a model uses to process input and output. A token can be:
The number of tokens in text is determined by the tokenizer used by the model. Refer to the usage statistics returned by the API for actual usage. Each model has a maximum input and output token count (see the model list). Exceeding this limit causes failure. |
Online inference | Refers to all direct and indirect invocations of a model, primarily covering the following scenarios:
|
Batch offline inference (Batch File) | Large-scale data processing performed offline for scenarios that do not require real-time responses, using the OpenAI-compatible - Batch (file input) API. |
Batch online inference (Batch Chat) | Batch online calls made through the Batch Chat API. |
FAQ
Q: How do I view the total token usage for my Alibaba Cloud account?A: Use your Alibaba Cloud account (primary account) to access the Bill Details page and export the bill, then view the token usage in the bill.
On the Bill Details page, filter and export as follows:
- Set the Billing Month to the target month (such as
2025-05). - Under Product Name, select Model Studio.
- Click the download icon in the upper-right corner of the table to export the bill data.
A: Not currently supported. The Alibaba Cloud app does not currently support viewing Model Studio free quota usage or model usage data. Log in on a PC and go to the Free Quota or Model Usage page in the Model Studio console.
Why do charts or details show "-" or empty values?
When you view data by model, API Key, workspace, or other dimensions, some values may appear as "-" or be empty. Common causes:
- A "-" may indicate that the model does not support the metric.
- Data for a specific dimension is delayed. Right after a call, the value may appear as "-" and change to the actual value after a few minutes or more.
- An empty API Key usually indicates that the call came from the console.
- For calls made through Batch Chat, API Key ID-related data is empty.
- Region information is empty for some models.
- The invocation method is empty for speech models.