Agent cost simulator
Model multi-turn tool loops the way agents actually burn tokens — sticky context, per-turn I/O, cache hits, and tasks per day. Rank every model by monthly cost. URL updates as you type so you can share a scenario.
FAQ
How is agent cost different from a chat calculator?
Agents call the model many times per task and re-send large system/tool context. This simulator multiplies turns × tasks and applies cache-read pricing on sticky context after the first turn when the model supports it.
What does cache hit % mean?
On turns 2+, that fraction of the sticky context is billed at the provider’s cache-read rate instead of full input. 85–95% is typical for coding agents with a stable system prompt and tool schemas.
Are reasoning tokens included?
Put estimated thinking/reasoning tokens in the “output / turn” field — most APIs bill them as output (or a similar completion rate). Adjust upward for high-reasoning modes.