Whenever you run a model, you’re creating a prediction.

Some models run very quickly and can return a result within a few milliseconds. Other models can take longer to run, especially generative models, like the kind that produce images from text prompts.

Prediction statuses

Predictions can have any of the following statuses:

• starting: the prediction is starting up. If this status lasts longer than a few seconds, then it’s typically because a new worker is being started to run the prediction. Refer to cold boots.
• processing: the predict() method of the model is currently running.
• succeeded: the prediction completed successfully.
• failed: the prediction encountered an error during processing.
• canceled: the prediction was canceled by the user or exceeded its deadline after starting.
• aborted: the prediction exceeded its deadline before it could start running.

Timeouts

Predictions time out after running for 30 minutes. If you require more than 30 minute timeouts for predictions, contact us.

You can also set a deadline to automatically cancel a prediction if it doesn’t complete within a specified duration. This is useful for building applications where you need to decide when to stop waiting and move on - for example, showing a timeout message to users or switching to an alternative workflow.

What happens when a prediction exceeds its deadline depends on whether it had started running:

• Before starting: The status changes to aborted
• After starting: The status changes to canceled

Billing for predictions with deadlines

You’re billed based on whether the prediction had started running:

• Aborted (never started): No charge
• Canceled (started running): You pay for the time the prediction ran before it was canceled

This means you only pay for compute resources that were actually used.

Monitoring predictions

When you’re logged in, you can view a list of your predictions on the dashboard, with summaries of status, run time, etc.

For long-running models, you may want to poll the API or use webhooks to check the status of a prediction.

To learn more about how long predictions are stored with Replicate, refer to Data retention.