A job that takes four attempts costs four times as much under token pricing, even though you only got one result. We think that is backwards.
Tokens hide the waste
Token bills measure effort, not outcome. Retries, abandoned plans and long dead ends all count the same as the work that shipped. Teams end up optimising prompts for cost instead of for results.
2.7xaverage attempts per finished job
38%lower spend after the switch
0charges for failed runs
Pricing per finished run
A run is one instruction carried through to a finished result, however many agents and attempts it takes. If it fails, you are not charged. That puts the cost of reliability on us, which is where it belongs.
- Starter: free up to 10,000 runs a month
- Scale: USD 49 a seat, plus 0.09 cents a run
- Enterprise: a volume rate above 500,000 runs
Budgets you set before the run
Every job can carry a budget. The runtime checks it before each step, so a run that is about to overspend stops and asks instead of finishing on your card.
- Pricing
- Cost control