OpenAI has updated prompt caching for the GPT-6 family to help persistent agents and long-running conversations reuse shared context more reliably. Eligible prompt prefixes reused within a 30-minute window can receive discounts of up to 90% on cached input tokens, while higher cache hit rates are enabled by default.
Developers also gain a Prompt Caching Dashboard, diagnostics for investigating unexpected misses and controls for deciding how much of a prompt should be cached. The practical benefit is lower latency and cost for applications that repeatedly carry forward instructions, tool definitions and conversation history.