OpenAI announced on September 22, 2026 an improved prompt caching system for its GPT-6 model family, according to a post on the company's website. Prompt caching reuses shared context—such as instructions, tool definitions, and earlier turns—across API requests to reduce response times and cost. OpenAI says it gives developers discounts of up to 90% on cached input tokens, with cache discounts applied for eligible shared prefixes reused within a 30-minute window. The update introduces a Prompt Caching Dashboard that shows how much of an application's input is served from cache and tracks hit rates over time, plus a diagnostics tool that compares a request with a recent response to identify changes to the model, tools, settings, or input that prevented reuse. The company also describes optional controls: explicit cache breakpoints to choose which prompt prefixes to reuse; the ability to change reasoning effort between responses without breaking cache by appending a configuration_update; guidance to preserve cache as tools and instructions change using allowed_tools, tool_choice, and developer messages; and prewarming to prepare known context ahead of time. OpenAI frames the changes as support for persistent agents that work for hours on tasks such as refactoring codebases or producing documents. The announcement is a first-party vendor post; independent verification is limited.
- Discounts of up to 90% on cached input tokens, per OpenAI
- Cache discounts apply to eligible shared prefixes reused within a 30-minute window
- New Prompt Caching Dashboard and diagnostics tool for monitoring and misses
- Reasoning effort can now be changed between responses without breaking cache
- Announced September 22, 2026
What it means for you
If you or your team build software on top of OpenAI's API—especially agents that run for a long time and reuse the same instructions repeatedly—these updates can make those requests faster and cheaper. Caching means OpenAI reuses work it already did instead of charging you full price again, and the new dashboard and tools help you see whether that's actually happening. If you only use ChatGPT through the website or a no-code tool, this changes nothing for you.
Try this
If your team calls the OpenAI API directly, open the new Prompt Caching Dashboard and check your cache hit rate for a heavy workflow—if it's low, run the diagnostics tool to see what's breaking reuse.
Who should care
Developers and technical teams building GPT-6 API applications, particularly long-running agents that carry forward shared context across many requests.
Skip this if
You use ChatGPT or AI tools through a web interface or no-code platform and don't write code against the OpenAI API.
Sources: OpenAI — read the original