How we price runs, and why it isn’t per token

When we started, we priced like everyone else: per token. It took three months to see the problem.

Per token punished good habits

The teams that wrote careful prompts, added verification steps and ran evals used more tokens. They paid more for being responsible.

What a run is

A run is one end-to-end task: an agent receives an input and finishes, fails or hands off. Retries, tool calls and verification inside it are included.

What we learned from the switch

• Customers added verification steps once they stopped costing extra.

• Support questions about bills dropped by two thirds.

• Forecasting became simple: runs track business volume, tokens don’t.

Limits that protect you

Every plan has a spend cap per run, so a looping agent stops instead of running up a bill.

Pricing should reward the behaviour you want customers to have.