OpenAI announces advanced caching system for GPT-6
The new system offers developers higher cache hit rates, a thirty-minute reuse window, and new monitoring tools.
OpenAI has announced a new prompt caching system developed for the GPT-6 family, implementing updates that reduce costs and improve response times in shared contexts.
GPT-6 and Complex Tasks
GPT-6 enables the operation of persistent agents working for hours on complex tasks, ranging from refactoring codebases to preparing comprehensive research and presentations.
API Requests and Caching
Applications behind these agents typically make a series of API requests built upon one another, carrying the same instructions, tool definitions, and context from previous turns.
OpenAI caches this shared context to reuse computation between requests, which shortens response times.
Cost Advantages and Discounts
The system significantly reduces operational costs by offering developers discounts of up to 90 percent on cached input tokens.
New System and Tools
The advanced caching system provides higher hit rates by default, a 30-minute reuse window, and new monitoring and diagnostic tools.