Better prompt caching for GPT-6
OpenAI improved GPT-6 prompt caching with higher default hit rates, a 30-minute discount, and diagnostics.
OpenAI says GPT-6 prompt caching now delivers higher hit rates by default and discounts eligible shared prefixes reused within a 30-minute window, up to 90 percent off cached input tokens. A Prompt Caching Dashboard and diagnostics tool show hit rates and why a miss occurred, such as changed tools. Developers can set cache breakpoints, change reasoning effort without breaking cache, keep tool definitions stable, and prewarm shared context to reduce latency.
- Eligible shared prefixes reused within 30 minutes receive cache discounts.
- A dashboard and diagnostics tool show hit rates and miss reasons.
- Reasoning effort can change between responses without invalidating cache.
- Breakpoints, stable tool definitions, and prewarming are optional controls.
- Vendors
- OpenAI
- Products
- Prompt Caching Dashboard
- Organizations
- OpenAI
- AI models
- GPT-6
From the original article
3 min read at openai.com
Learn how GPT-6 improves prompt caching with higher cache hit rates, new diagnostics, explicit breakpoints, and controls that reduce latency and costs.
openai.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.