How much do batch APIs discount?

Both OpenAI and Anthropic offer roughly a 50 percent discount on input and output tokens for batch jobs in exchange for asynchronous, best-effort completion within a window. The discount is editable here so you can model other providers.

What is the trade-off with batch processing?

Batch jobs run asynchronously and can take up to 24 hours to complete, so they suit non-interactive workloads like bulk classification, evaluation, or embedding generation rather than user-facing chat.

Why is the time-to-completion only an estimate?

Batch completion time depends on provider queue depth and your job size. The calculator shows a simple proportional estimate against a typical 24-hour window, not a guarantee.

Is anything sent to a server?

No. The estimator runs entirely in your browser. You only enter counts and settings, and nothing is uploaded, stored, or logged.

Batch API Cost Estimator

Batch API cost estimator

If your workload does not need an instant response — bulk classification, offline evaluation, document tagging, synthetic data generation — the batch API can cut your bill roughly in half. This estimator shows the real-time cost of a job, the discounted batch cost, and the dollars you save by trading latency for a lower price.

How it works

The calculator multiplies your request count by the average prompt and completion tokens to get total input and output tokens. It prices those at the selected model’s real-time rates, then applies the batch discount (50 percent by default, editable) to produce the batch total. The difference is your saving.

It also shows a rough time to completion: batch jobs are best-effort within a provider window (commonly up to 24 hours), so larger jobs trend toward the far end of that window. Treat this as guidance, not a service guarantee.

Tips and notes

Batch suits asynchronous work only. Anything a user waits on in real time should stay on the synchronous endpoint.
Stack discounts where possible. On long shared prefixes, prompt caching and batch pricing can both apply — model caching separately with the caching-savings calculator.
Prices are editable estimates. Providers change rates and discount terms; confirm current batch pricing in your provider dashboard before budgeting.