OpenAI Upgrades GPT-6 API Prompt Caching to Cut Agent Latency and Input Costs

OpenAI has upgraded prompt caching for GPT-6 in the API, aiming to raise cache-hit rates by default so long-running agents can reuse more context, respond faster and qualify for cached-input discounts of up to 90%. The September 22 update also adds monitoring, diagnostics and controls for developers who need to understand why cache misses happen and where reusable prompt prefixes should end.