Unit economics, not vibes

What does an AI agent actually cost to run?

Every agent product is a margin bet on token prices. Pick a model, set your workload shape and your pricing, and see where the gross margin lands, where it breaks, and how every major model compares on your exact workload.

WORKLOADS

Assumptions

Model

Input $ / M tok

Output $ / M tok

Input tokens per task

Output tokens per task

Input served from cache

Tasks per user per month

Price per user per month ($)

The economics

$0.00 model cost per task

Fresh input Cached input Output

COGS / user / mo

Gross margin

HEALTHY

Break-even tasks / user

Margin vs. usage

Flat subscription pricing means every extra task eats margin. The marker is your current assumption.

Every model, your workload

Same tokens, same pricing, every model. Sorted by cost per task.

Model	$ in / $ out	Cost / task	COGS / user	Margin