AnalysisAI ModelsJuly 10, 2026

GPT-5.6 prompt caching appears broken

Users report that prompt caching on GPT-5.6 charges the new 1.25× cache-write fee but never credits cached_tokens. Reproducible with simple curl requests.

1 source

GPT-5.6 prompt caching appears completely broken: you pay the new 1.25× cache-write fee, but cached_tokens never credits — reproducible with two curlsreddit.com

Back to the feed