Today

SourceVane · Practical AI

AI API cost calculator

Compare two AI API budgets using token prices, retries and successful tasks. Enter your own assumptions; no API calls or account required.

Enter your workload

First comparison? Keep the sample values, then select Compare monthly costs.

The starting numbers are illustrative assumptions, not current provider prices or measured success rates. For a real budget, replace both options with your own usage and the exact provider’s current rates.

  1. Count original tasks per month, not individual calls.
  2. For each option, enter tokens per billed call and the average number of billed calls per original task. For example, 110 calls across 100 tasks means 1.1 attempts.
  3. Enter the percentage of original tasks accepted after all retries. For example, 95 accepted out of 100 means 95%, not 0.95.

Zero tasks, tokens or prices are allowed. Attempts must be at least 1. With zero tasks or 0% success, cost per successful task cannot be calculated.

Choose a verified standard text price or enter your own rate. Each option can have different token use, retries and final success. Calculations stay in your browser.

Catalogue presets are temporarily disabled. No current price catalogue has passed verification in this environment. Enter a current rate from the provider’s official page.
Workload
Option A
Option B

Submit your inputs to see both options.

Share these assumptions

The link contains the numeric values shown in the form. It is not encrypted, and it does not store the values on SourceVane’s server.

How the estimate works

For each option, monthly cost = tasks × that option’s average attempts × (input tokens × input price + output tokens × output price) ÷ 1,000,000.

Cost per successful task = monthly cost ÷ (tasks × final success rate). Count success after the allowed retries, using your own acceptance test. cost-v2 keeps each option’s token use and retry rate separate.

Use effective billed rates for your configuration. Cache writes and reads, batch discounts, tool charges, hosting, taxes and subscriptions are not added separately. This estimates token spend, not latency, model quality or a guaranteed invoice.

Compare the estimate with measured usage before budgeting. Measure retry cost · Compare batching and caching.

Explore API cost decisions · Review the retry-cost study · Follow future guides

Privacy choices

This version uses no advertising or audience analytics. Theme preferences stay on your device. Calculator values stay in your browser.