Batch AI Cost Savings Calculator

Many providers offer a cheaper batch or offline mode for work that does not need an immediate answer. Enter the tokens per job and the number of jobs, then either the batch prices or the discount percentages, and compare it with standard pricing: standard total, batch total, savings, percentage and cost per job. The discount is never assumed: it is whatever you enter, or a dated preset read from the provider.

Processed locally in your browser. Your data is not uploaded to our servers.

Average prompt size of one job.

Average reply size of one job.

The total batch to price.

A preset fills in the standard and batch prices and switches to “enter prices”. You can edit them.

Add it to project the savings per day and per month.

30 by default.

Preset prices are standard and batch prices in USD per million tokens, read from each provider’s own pricing page and last verified 2026-10-19. Providers set their own batch discounts and may change them, and batch jobs usually complete within a longer window: confirm on the provider’s page.

Price sources:

Result

Enter the tokens per job, the number of jobs and the prices to compare.

Exact arithmetic on your prices. Latency, completion windows, rate limits and which workloads qualify for batch pricing are outside this comparison.

Examples

Worked examples (example prices of 3 input and 15 output per million tokens, with batch at half price):

WorkloadJobsStandard totalBatch totalSavings
Classifying support tickets50,000105.00 USD52.50 USD52.50 USD
Summarising documents5,000157.50 USD78.75 USD78.75 USD
Offline model evaluation200,0001,800.00 USD900.00 USD900.00 USD

Formula & Methodology

The arithmetic is exact for the numbers you enter:

cost per job = input tokens × input price ÷ 1,000,000 + output tokens × output price ÷ 1,000,000standard total = cost per job at standard prices × jobs; batch total = cost per job at batch prices × jobssavings = standard total − batch total; savings % = savings ÷ standard totalbatch price from a discount = standard price × (1 − discount %)

Good to know

  • There is no universal batch discount. Each provider sets its own, per model, and may change it. Enter the figure from your provider’s pricing page, or use a dated preset.
  • Batch work is usually asynchronous: results arrive later, within a window the provider sets, and some rate limits and features differ. The time cost is not modelled here.
  • Not every request qualifies. Streaming, tool use or interactive features may not be available in batch mode.
  • Batch pricing and prompt caching can sometimes be combined, but the rules differ by provider. Model caching separately with the prompt cache calculator.
  • Preset prices are conveniences, dated and sourced. Use your own prices for negotiated rates or newer models.

How to Use

  1. Enter the input and output tokens per job and the number of jobs.
  2. Enter the standard prices, then the batch prices or a discount percentage, or prefill them from a preset.
  3. Optionally enter the jobs per day to project a month.
  4. Read the standard and batch totals, the savings and the cost per job.

Frequently Asked Questions

How much cheaper is batch AI processing?

It depends on the provider and model: each sets its own batch prices. Enter the batch prices or the discount percentage from the provider’s pricing page and this calculator shows the total savings for your workload.

Is there a standard batch discount I can assume?

No. Some providers currently publish a flat percentage and others price by model, and either can change. That is why this tool never assumes one and asks for your numbers, or a dated preset.

What do I give up with batch processing?

Speed. Batch jobs are processed asynchronously and finish within a longer window, and some interactive features may be unavailable. This calculator compares cost only, so decide whether the delay suits the work.

Can I use the discount percentage instead of prices?

Yes. Switch the mode to a discount percentage, enter the standard prices and the percentage off input and output, and the batch prices are worked out for you.

Does this include prompt caching?

No. It compares standard and batch prices for the tokens you enter. To model caching, use the prompt cache savings calculator.

Last updated: