Latest AI news · Original source attached
OpenAI overhauls prompt caching for GPT-6
Higher cache hit rates, up to 90% discounts on cached tokens, and new dashboards and diagnostics for builders of persistent agents.
Reported from OpenAI
Summary written from the original first-party source — the link to it is at the end of this post.
OpenAI shipped an improved prompt caching system for the GPT-6 family, delivering higher cache hit rates by default, discounts of up to 90% on cached input tokens, and new tools to monitor and diagnose cache performance. Reused prefixes stay eligible within a 30-minute window.
What’s actually new
- Higher hit rates by default. Shared prefixes reused within the window get discounts of up to 90%, so persistent agents running for hours keep more context cheaply.
- Visibility. A new Prompt Caching Dashboard shows hit rates and token composition over time, and a diagnostics tool explains unexpected cache misses and their token impact.
- More control. Explicit cache breakpoints let developers choose which prefixes to cache, reasoning effort can change between responses without breaking cache, and prewarming prepares known context ahead of user requests.
Why it matters
Caching is the economics of long-running agents; it decides whether hours-long coding and research tasks stay viable. GitHub says these changes cut the share of prompt tokens needing fresh processing by more than 50% across billions of requests. For developers building on GPT-6, smarter caching directly lowers the cost of agentic workloads.
(Original source: OpenAI)
AI Slop Weekly
Get this story in context, every Sunday.
The monthly rundown of what the stories above add up to — what shipped, what changed, what to actually use. 3-minute read, free forever.
Subscribe freeNo spam. Unsubscribe anytime.