The update adds longer cache reuse, diagnostics and new controls designed to cut latency and costs for agents running across extended tasks.
OpenAI has upgraded prompt caching across the GPT-6 family, introducing higher cache hit rates and new controls aimed at making persistent AI agents faster and cheaper to operate.
Prompt caching lets applications reuse computation when multiple API requests share the same instructions, tools or context. Cached input tokens can receive discounts of up to 90%, while GPT-5.6 and later models can reuse eligible cached prefixes for at least 30 minutes after their most recent use.
The improvements are particularly aimed at agents that work for extended periods and repeatedly carry large amounts of context between requests, such as coding agents refactoring a codebase or systems producing long research documents.
OpenAI has also launched a Prompt Caching Dashboard that lets developers track cache hit rates and compare cached and uncached token usage. A separate diagnostics tool can identify changes to models, tools, settings or inputs that caused a request to miss the cache.
Developers can now set explicit cache breakpoints to control which portions of a prompt are reused. GPT-6 models also allow reasoning effort to change between responses without invalidating previously cached context when configured appropriately.
OpenAI is also adding cache prewarming, which lets applications process shared instructions, tool definitions or reference material before a user sends a request, reducing the amount of processing required before the model begins responding.
GitHub said improvements to OpenAI’s caching infrastructure reduced the share of prompt tokens requiring fresh processing by more than 50% across billions of requests to OpenAI models.
The changes expand on prompt caching introduced by OpenAI in 2024, which initially offered automatic discounts for repeatedly used prompt prefixes. The GPT-6 system extends those capabilities with longer reuse windows and more direct developer control over cache behavior.
