Open History to see the last commit
- replicate: cacheRead — Replicate bills language models per input token and per output token only; no model's billing config or the pricing page states a cached-input price, https://replicate.com/pricing, checked 2026-10-06
- replicate: requestMap — not applicable rather than unpublished: Replicate publishes each model's request fields, but under the prediction's
inputobject ("The input schema depends on what model you are running"), whichChatRequestMap(an OpenAI-style top-level body) cannot express, https://api.replicate.com/openapi.json, checked 2026-10-06