OpenAI Improves GPT-6 Prompt Caching With Up to 90% Savings and New Diagnostics

OpenAI upgraded prompt caching for GPT-6 models with cache discounts for shared prefixes reused within a 30-minute window, up to 90% savings on cached input tokens, a new caching dashboard, explicit cache breakpoints, and mid-conversation reasoning-effort changes that preserve the cache.