Question 1

Where does this pricing data come from?

Accepted Answer

Public list pricing per 1M tokens for common model families. Providers change rates often, so treat the output as a directional estimate, not a quote.

Question 2

Why does agent loop count matter for cost?

Accepted Answer

Every additional loop iteration re-sends context and generates new output, multiplying token cost per request. An agent that loops 5 times costs roughly 5x the tokens of a single-pass call.

Question 3

What counts as a 'retrieval call'?

Accepted Answer

Any vector database or search query made to fetch context before generating a response. Each one adds both a query fee and the tokens of the context it injects.

Question 4

How is the 'potential optimization' number calculated?

Accepted Answer

It flags specific inefficiencies — excess retries, unnecessary loops, uncached retrieval or tool calls, an oversized model, or over-provisioned infrastructure — and estimates the savings from fixing each one, capped at 50% of current cost.

Question 5

Why show cost at 10x usage?

Accepted Answer

Token and infrastructure costs don't always scale linearly. Seeing the 10x number early avoids getting surprised by a bill once usage grows.

AI Cost Calculator

About the Assessment