Atlas · models
Cost and latency
Why long sessions get slower and more expensive, and what caching changes.
See it working: Plate I
In plain words
Every turn sends the whole context again. Long sessions mean more tokens per turn and more turns.
How it works
Providers publish prices per million input and output tokens. Prompt caching can reuse processing for a repeated prefix, so repeated input is not always charged or delayed the same way.
Going deeper
Plate I’s tokens-read counter adds up raw context size per turn. A real bill differs because of caching and output pricing, so check a provider’s live price table rather than a number frozen on a page.
Sources
- Prompt caching OpenAI
Prompt caching reuses work when requests share the same prompt prefix.
- API pricing OpenAI
Prices per 1M tokens.
Quotes were checked against the live pages on 2026-11-08.