LLM tokens and cost
Batch API discount calculator.
Enter the workload, your prices and the batch discount. Work that cannot wait for an asynchronous batch stays at the standard price.
Batch cost = batchable cost x (1 - discount) + everything else at the standard price. 100,000 requests at 3,000 input and 800 output tokens, with example prices of $2.00 and $8.00 per 1M, cost $1,240.00 at the standard price and $620.00 at a 50% batch discount, saving $620.00.
Your numbers
Result
Worked example
100,000 requests with 3,000 input and 800 output tokens, example prices of $2.00 and $8.00 per 1M, and a 50% batch discount. One request: (3,000 x 2 + 800 x 8) / 1,000,000 = $0.0124. All 100,000 at the standard price: $1,240.00. All batched: $1,240.00 x 0.5 = $620.00, a saving of $620.00. If only 60% of the work can be batched, the batchable $744.00 falls to $372.00, the rest stays at $496.00, the total is $868.00 and the effective discount is 30.0%.
How it works
OpenAI's Batch API documentation describes asynchronous groups of requests with a 50% cost discount against synchronous APIs and a 24-hour turnaround window. Anthropic's Message Batches documentation says all batch usage is charged at 50% of the standard API prices. Those are the documented figures at the time of opening; they can change and can differ by model, so the discount is an input.
The trade is latency for price: a batch result can take hours, so only work nobody is waiting on (evaluations, bulk classification, offline summaries, backfills) belongs in it. Anthropic's pricing page says its prompt caching multipliers stack with the Batch API discount; this page does not model the combination, so use the prompt caching calculator separately and compare.
Sources and basis (opened 2026-10-08): Batch API | OpenAI API; Batch processing - Claude Platform Docs; Pricing - Claude Platform Docs. Everything else is arithmetic on your inputs.
All prices on this page are examples you can overwrite, not any provider's current prices. Check your provider's pricing page before you rely on a number. Results are planning estimates, not a quote or financial advice. Everything runs in your browser. Nothing you type is sent or stored.
FAQ
How much does the Batch API save?
OpenAI and Anthropic both document a 50% discount for batch requests against standard prices. In this example that turns $1,240.00 into $620.00.
Can every request use a batch?
No. Batches are asynchronous, so any request a user is waiting on has to use the standard API. Enter the share that can wait and the calculator discounts only that part.
How long does a batch take?
OpenAI documents a 24-hour completion window and says batches are often faster. Check your provider's documentation for its window and for limits on batch size.
Can batch and prompt caching discounts stack?
Anthropic's pricing page says its prompt caching multipliers stack with the Batch API discount. This calculator does not combine them, so work out each separately and check the provider's pricing page for how the multipliers interact.