OpenAI improves prompt caching for GPT-6
OpenAI prompt caching cuts GPT-6 costs and speeds up persistent agents Reuse shared context for up to 90% savings and better cache hits
OpenAI says it has updated prompt caching for GPT6 to improve cache hit rates and help persistent agents run faster and at lower cost. The company says cached input tokens can receive discounts of up to 90% when shared context is reused across requests.
The new system applies cache discounts to eligible shared prefixes reused within a 30minute window. OpenAI also introduced tools for monitoring cache performance, diagnosing misses, and selecting how much of a prompt to cache.
The company highlighted ways developers can optimize caching, including using explicit cache breakpoints, keeping tool definitions stable, changing reasoning effort without breaking cache, and prewarming known context before users make requests. OpenAI says these changes are meant to improve performance for longrunning agent workflows.