OpenAI gives GPT-6 cache diagnostics and explicit breakpoints

single source· 1 articles · confidence: low · first seen 2026-09-22 21:00 UTC

What this means for you

If you cache long prefixes, the diagnostics are the part to check: they show whether requests you assumed were hitting the cache were not. OpenAI published no hit rate and no price change, so you cannot put a number on the saving yet.

OpenAI has changed how prompt caching works for GPT-6. Prompt caching reuses the opening stretch of a request — a system message, a long document — instead of processing it again on every call, which makes repeat requests cheaper and faster. The update adds diagnostics showing why a request missed the cache, explicit breakpoints marking where a cached stretch ends, and other controls. OpenAI says the result is more cache hits and lower latency and cost. No figures were published: not a hit rate, not a price change.

Key facts

  • ·OpenAI's post on prompt caching for GPT-6 was published on 22 September 2026. source
  • ·The update adds diagnostics for prompt caching, which the post says show why requests miss the cache. source
  • ·The release introduces explicit breakpoints, markers for where a cached stretch of a prompt ends. source
  • ·OpenAI says the changes raise cache hit rates and reduce latency and cost. source
  • ·The post gives no hit-rate figure, no price change and no benchmark measurement. source

What the sources say

  • OpenAI News — Vendor post describing higher cache hit rates, new miss diagnostics and explicit breakpoints, with no figures attached.

Sources

The original reporting. Follow these — they did the work.

← the wire