Paste an OpenAI usage export, then select Find cost anomalies. A sample export is loaded so you can try it right away.
About the OpenAI API cost anomaly card
The OpenAI API cost anomaly card turns a pasted usage export into a fast read of what drove your API spend. Paste the export, choose whether to group by model, project, endpoint, or day, and the tool normalizes cost, token, and request columns, aggregates spend by your chosen dimension, and flags the outliers that dominate the bill, each with a suggested owner action. It is built for the moment before a month-end spend review when you need to know which models and projects to look at first.
A sample export is loaded so you can see findings right away: one model and one project spike mid-month and surface as the spend anomalies. Everything runs in your browser. The export you paste is never uploaded, logged, or stored, which matters because a usage export can carry project and organization names, model names, and cost data. When the outliers look right, download the anomaly CSV or copy the budget-review note. A dollar figure is shown only when your export includes a cost column, so the tool never asserts model pricing of its own.
How to use
- In the OpenAI platform, open the Usage dashboard and export your usage for the period under review as a CSV (or copy the rows). Include a cost column and, where available, model, project, endpoint, request, and token columns.
- Paste the export into the usage box. A sample export is loaded so you can see the format the tool expects.
- Choose how to group the anomalies: by model, project, endpoint, or day.
- Select Find cost anomalies to see the spend outliers for your chosen grouping with token and request context and a suggested action.
- Download the anomaly CSV or copy the budget-review note for your spend review. If your export has no cost column, findings rank by token usage instead.
Worked examples
A mid-month model spike is flagged as the top spend driver
When one model's spend jumps mid-month and dominates the bill, it is flagged high severity with a suggestion to consider a cheaper tier, token caps, or caching.
A single project far above the others is surfaced even with a small share
A project whose spend is several times the median project is flagged on the ratio even when its share of the total is small, so a one-off heavy run does not hide behind a large total.
An even spread of spend shows no anomaly
When no single model, project, endpoint, or day dominates spend, the tool reports no clear cost anomaly and lists the top groups for reference, so you can close the review with confidence.
Frequently asked questions
- What should I paste in the usage box?
- An OpenAI usage export as CSV, TSV, or JSON rows. The tool reads a cost (USD) column as the spend driver, and uses model, project, endpoint, request-count, and token columns when present. It recognizes common header names (for example cost, amount, model, snapshot_id, project, n_requests, n_context_tokens, n_generated_tokens, total_tokens). You can omit any columns you do not have.
- How does it decide which groups are an anomaly?
- A group is flagged when it is both a large share of total spend and well above the typical (median) group, or when it is several times the median on its own. That two-part rule means an even spread of similarly sized groups is not falsely flagged, while a single dominant model, project, endpoint, or day, or a one-off heavy run, is.
- Why is there no dollar amount sometimes?
- The tool shows dollars only when your export includes a cost column. Without one, it ranks findings by token usage and tells you so, rather than inventing a price. It never asserts model pricing of its own, because OpenAI pricing changes over time and varies by model; always verify against your billing.
- Is my export sent anywhere?
- No. Parsing, grouping, and outlier scoring run entirely in your browser. The pasted rows, project and organization names, model names, endpoints, request counts, token counts, and cost data are never uploaded, logged, or stored, and they are not included in any analytics. Only coarse, anonymous count bands are recorded so we can tell how often the tool is used.
- Does this replace OpenAI usage limits or a FinOps platform?
- No. It is a fast, paste-based triage aid for a spend review. It does not connect to your account, set budgets, or track cost over time. Keep using OpenAI usage limits and budgets, the billing dashboard, and a platform like Vantage, CloudZero, or Finout for authoritative, ongoing cost management.
- Is the OpenAI API cost anomaly card free?
- Yes. It is free to use and does not require an account.
Use this again tomorrow
Save this page so it's one tap away when you need a quick result.
Take a 2-minute brain break.
Play Daily Challenge on sts.games