Gengxi API Developer Docs

Prompt caching

Understand when stable repeated context may reduce repeated input cost.

How prompt caching works

Stable repeated context may be cached by supported upstream models and routes, which can reduce repeated input costs.

What affects a cache hit

  • Model
  • Provider
  • Account type
  • Request path
  • Context stability

Design for stable prefixes

Keep reusable instructions and large shared context stable when the selected model and route support caching. Put frequently changing user content after the stable context where the protocol permits.

No cache guarantee