Prompt caching
Reuse **identical prefix tokens** — same system prompt or RAG block every request? Cache it.
1Try it yourself
Playground
Prompt cache lab
Reuse identical prefix tokens — cut cost and latency when system prompt or RAG block repeats.
Same system prompt every request
2Read & reflect
Recap
#What you just did
You picked cache vs skip for repeated vs one-off prompts.
Teach
#How it works
Cache-friendly patterns:
- Stable [system prompt](/reference/system-prompt) at the start
- Fixed RAG context block before user message
- Same tool schemas every call
Changing the prefix busts the cache. Put volatile user text after the static block.
Use it
#When you'd use this
- High-volume APIs with identical instructions
- RAG apps with shared doc bundles per tenant
Watch out
#Watch out
Cache discounts vary by provider. Measure — don't assume every prefix qualifies.
Try next
#Try this next
Reorder one prompt: static system + docs first, user question last.
3Spark check