Overview
You can review API usage in two ways:
Use Analytics > Usage to review API activity across current and past reporting periods.
Read the usage information in an API response to check an individual request.
Playground uses the same API calls as your applications. Its tokens count toward your account’s API usage and follow the same usage rules and pricing.
Dashboard access
You must be an organization owner or have the Usage Dashboard permission to access the Usage Dashboard.
Review usage in the dashboard
In the API Platform, open Analytics and select Usage to review the current project. For organization-wide reporting, select Organization analytics. Select the reporting period you want to review, and check it again after switching between project and organization analytics.
Dashboard data is displayed in UTC. Keep that time zone in mind when comparing dashboard activity with your application logs.
Recent usage may take time to appear in the dashboard. If recent activity is missing, check again later.
Filter usage
In project Analytics, usage is scoped to the selected project. In Organization analytics, use Project to select the projects you want to review. All projects includes every project in the organization, subject to any other filters you have selected.
Use Filters to narrow usage by API sources, API keys, Models, Users, or Processing tier. API sources lets you select All API sources, Agents API, or Direct API.
Remove individual selections or select Clear filters to reset the Filters panel. Clearing filters does not switch between project and organization analytics.
Use Filter on the Usage chart to filter by Batch. Select Reset chart filters to reset that chart’s filters.
If a capability does not support the selected filters, its values may be unavailable. Unavailable values do not mean zero usage.
If a filter displays a permission error, you may not have access to its options. This does not mean that no usage was recorded.
Review an individual user’s usage
Open Usage in the project or organization Analytics view you want to review. In the API capabilities table, select View for Responses and chat completions.
In Filters, select Users and select the user. The results update immediately.
Review the chart and table for that user.
To review a user across capabilities, apply the Users filter on the Usage overview. To compare users, clear that filter and select the Users table tab.
View tokens per day, hour, or minute
Where interval controls are available in Usage, select 1d or 1h to view tokens by day or hour, or 1m to examine tokens per minute (TPM). Select Auto to let the dashboard select the time interval for your reporting period. These intervals apply to usage data, not billing data.
If 1m is disabled, select a shorter reporting period or use Select to zoom and drag across the chart to narrow the range. Then select 1m. Press Esc to cancel a zoom selection. Select the current date range in the Analytics toolbar to choose a different reporting period.
Review usage charts and tables
Use Metrics to select one or more series to display together, including Input tokens, Input uncached tokens, Input cached tokens, Cache write input tokens, Input audio tokens, Output tokens, and Output audio tokens. Available metrics depend on the capability. Use the legend to identify each series.
Use Group by to organize usage by categories such as Project, User, or Model, where available. Filters control which usage is included; grouping controls how it is displayed. Switch between line and bar charts to inspect trends or compare categories.
Use the Usage table tabs to review the available breakdowns. Tables show the relevant measures for each row, such as requests, input and output tokens, or capability-specific units.
In API capabilities, select View for a capability to open its detailed charts and breakdowns. Responses and chat completions includes separate Tokens and Requests charts and a Model requests metric. Its breakdown tabs use the same dimensions as the overview, without the API capabilities tab. Use the Usage breadcrumb to return to the overview.
After selecting a Group by option, the chart may combine additional categories into Other. If Other appears in the chart legend, select its label to see the included categories and their usage values.
Select Compare dashboards to view Usage alongside another dashboard for the same reporting period. Use Left dashboard and Right dashboard to select the views. If the selected time interval is unavailable for the dashboard pair, the view switches to Auto. Select Exit comparison to return to the dashboard.
Investigate usage you don’t recognize
If Usage shows activity you can’t match to requests in your application logs, use these checks:
Confirm the organization, project scope, and reporting period. Use UTC when comparing dashboard timestamps with your application logs.
Use the project view for the affected project, or select it under Project in Organization analytics. Then use Filters to narrow usage by API key and model and select a shorter reporting period.
Compare the activity with logs for the same period, project, and API key. Check whether your logs cover all applications or integrations using that key.
If several applications or integrations share the same project or API key, the total alone may not identify which one sent a request. Compare the dashboard with records from each application for the same UTC period.
If you still can’t account for the usage, contact OpenAI Support with the affected UTC period, organization and project IDs, the API-key identifier, and relevant screenshots or logs. Include request IDs if available. Do not include secret API keys.
Export usage data
To export usage, select the Export CSV icon in the Analytics toolbar, then select Activity data.
Select the API capability and one or more Group by dimensions.
Review API Source, Projects, Users, API Keys, and Models. Projects and API Source are preselected from Analytics; select the other export filters in the dialog.
Select File format, Start time, End time, and Time interval, then select Download. Dates include full UTC days.
Activity and cost data are available to view and export from 00:00 UTC one calendar year ago through today. For older data, contact OpenAI Support. Daily data is exported in files covering 31-day increments.
For monthly invoice reconciliation and line-item exports, see: Exporting monthly usage details.
If an Activity export includes negative rows, compare rows with the same unit, grouping and dates. Use Cost data for billed amounts. A negative activity row alone doesn’t establish that your invoice was credited.
Review spend
Open Analytics and select Spend to review costs for the current project. For organization-wide costs, select Organization analytics and then Spend.
Check the current date range and the filters before comparing costs with usage. Usage and Spend offer different filters; a filter available in one dashboard may not apply to the other.
Use API sources, API keys, Models, and Line items to narrow spend. In organization analytics, use Project to include the projects you want to review. Use the Line items, API keys, Users, Projects, and Models table tabs for the available breakdowns.
An unavailable spend-limit value does not mean your limit is zero or that a new limit has been applied. Access permissions, an unset limit, or a failed request can prevent the dashboard from showing a value.
Review service health
If your organization has access to Service health, open that dashboard in Analytics to review API performance. For eligibility and metric definitions, see: Troubleshooting API errors and latency.
View credit grants
Credit-grant details are available on your organization’s Billing page.
Credit grants and API usage answer different questions: credit grants show the credits associated with your organization, while usage shows its API activity.
Review Scale Tier usage and costs
If your organization has Scale Tier enabled and you have the required access, open the Scale Tier usage page for detailed usage.
Scale Tier bundle costs are attributed to the organization, not to individual projects.
Usage is greater than zero, but project spend is zero
Check whether the model’s usage is covered by your Scale Tier allocation. If all usage for that model is within the Scale Tier TPM allocation, the project can show usage without additional model spend.
The Scale Tier subscription cost remains at the organization level. Review organization-wide spend rather than relying on an individual project’s spend.
Review multiple organizations
The Usage Dashboard does not combine cost or usage data across organizations. Each organization, including a sub-organization, is treated separately.
To view a sub-organization, select it in the organization selector.
If you need combined reporting, consider using projects within one organization instead of separate sub-organizations, or use the Usage API for custom analysis.
Check token usage in an API response
Token-usage field names depend on the endpoint.
| Token count | Chat Completions | Responses |
|---|---|---|
| Input tokens | usage.prompt_tokens | usage.input_tokens |
| Output tokens | usage.completion_tokens | usage.output_tokens |
| Total tokens | usage.total_tokens | usage.total_tokens |
Chat Completions responses can include token usage information. Examples include:
{
"usage": {
"prompt_tokens": 13,
"completion_tokens": 7,
"total_tokens": 20
}
}Depending on the endpoint and model, usage details can also include cached-input and reasoning-token counts. These help explain why the visible length of a response does not show its full token usage.
For an explanation of the different token categories, see: Understanding and counting tokens.
For a transcription response reporting token usage, audio and text input tokens are already included in input_tokens. Don’t add those components to the input total again. Compare output_tokens separately.
Get usage from streamed Chat Completions
For a streaming Chat Completions request, include:
{
"stream": true,
"stream_options": {
"include_usage": true
}
}With include_usage enabled, an additional chunk before data: [DONE] contains the usage for the entire request. Its choices array is empty. Other chunks have a null usage value.
If the stream is interrupted, you may not receive that final usage chunk. A missing final chunk does not establish that the request used no tokens.
This configuration is specific to Chat Completions. For details, see: Chat Completions API reference.
