GPT-6 Prompt Caching Adds Visibility, Diagnostics and Control

GPT-6 enables persistent agents to work for hours on complex tasks, while their applications make successive API requests that often carry forward instructions, tool definitions and context from earlier turns. [1] OpenAI says it caches this shared context to reuse computation across requests, reducing response times and providing developer discounts of up to 90% on cached input tokens. [1]

With the GPT-6 family, OpenAI launched an improved prompt-caching system that provides higher cache-hit rates by default and discounts eligible shared prefixes reused within a 30-minute window. [1] The new Prompt Caching Dashboard displays how much application input is served from cache, tracks hit rates over time and compares cached with uncached tokens. [1]

A diagnostics tool compares a request with a recent response to identify changes in the model, tools, settings or input that prevented reuse, while explicit cache breakpoints let developers select which prompt prefixes to reuse. [1]

Prompt caching is becoming an operational control surface for long-running GPT-6 agents, not merely an automatic discount. The dashboard makes cache behavior visible, diagnostics identify why reuse failed, and breakpoints put prefix selection in developers’ hands. Together, those controls let teams monitor and tune repeated instructions, tool definitions and context against response time and cached-token costs. The practical test is whether applications can preserve stable shared prefixes without constraining the changing context that complex tasks require. What the sources don't address: how cache-hit improvements vary across workloads or how much developers typically save in production.