Tool
AI model cost per finished task
Compare list price with the cost of a task that actually finishes, and the daily volume where self-hosting beats the API.
Attempts per task = 1 + average retries. Token cost uses each attempt. Review cost applies to the share of tasks that fail. Cost per successful task divides that total by the success rate. Listed cost is one attempt, before retries, failures, and review. A month in the break-even chart is 30 days. You type the prices. This page does not call a model.
Showing the example. Replace it with your prices.
Output
The comparison appears here.
How to use this
- Replace the two example models with the prices and token counts you were quoted. Add a third or fourth if you have them.
- Use the success rate and retries you actually see. A low list price can lose once failed tasks and review time are in the cost.
- Fill the self-host lines to see the daily token volume where the server costs less than the API, and whether your throughput reaches it.