OpenAI Launches Improved Prompt Caching for GPT-6

OpenAI has launched an improved prompt caching system for the GPT-6 family, designed to give developers discounts of up to 90% on cached input tokens. As detailed in OpenAI's announcement, the update delivers higher cache hit rates by default and provides cache discounts for eligible shared prefixes reused within a 30-minute window.

The update introduces new developer tools to monitor performance and diagnose cache issues. The Prompt Caching Dashboard tracks hit rates over time and features an input composition chart comparing cached and uncached tokens. When unexpected cache misses occur, a diagnostics tool helps identify changes to models, tools, settings, or inputs. For example, a sample diagnostic output showed a cache miss affecting 5629 tokens due to changed tools.

Developers can use explicit cache breakpoints to choose which prompt prefixes to reuse. On GPT-6 models, users can now change reasoning effort between responses without breaking the cache by appending a configuration_update while leaving request-level reasoning effort unchanged.

To preserve cache when tool use needs change, developers are advised to keep tool definitions, schemas, and ordering stable. Options include using allowed_tools to make only relevant tools callable, or setting tool_choice to none when no tools are needed, alongside new developer messages to append instructions. Additionally, prewarming allows applications to prepare known context ahead of time during startup to reduce latency.

The new features are available now, with monitoring and diagnostics accessible via OpenAI's platform dashboards and guides.