Pricing, without the plot twist.
Every plan follows one simple rule: returning an existing cached answer costs nothing. Choose how you want to handle the requests that need something new.
Choose your access
No plan adds a fee to cache hits. Move up only when you need new generations, identity-based limits, or more advanced caching.
$0
Free, of course.
Receive an existing answer when your request matches one already in Cache. Nothing new is generated.
$0
From Cache. Upstream usage applies.
Bring an upstream provider key to reuse a matching answer or generate a new one when there is no match.
$0
With a Cache account.
Generate within the free allowance and get consistent limits tied to your key - not your IP address.
Prepaid
Credits that only move on a miss.
Load credits and use them for new generations. Cache adds no markup and charges nothing extra.
Advanced reasoning, encrypted response caching, and custom controls for teams operating at serious scale.
The whole model
| Plan | Cache hit | Cache miss | Limits follow | You pay Cache |
|---|---|---|---|---|
| Anonymous | Returned free | Not generated | No account | $0 |
| Anonymous BYOK | Returned free | Your provider runs it | Anonymous access | $0 |
| Free | Returned free | Uses free allowance | Cache key | $0 |
| Paid | Returned free | Uses prepaid credits | Cache key | Generation cost only |
| Pro | Returned free | Custom reasoning workflow | Custom policy | Custom |
Good questions
Yes. If Cache already has an answer for the same request, it returns that answer without charging for another generation.
No. BYOK misses are billed by your upstream provider. Paid-plan credits cover the underlying generation cost, without an additional Cache fee.
Your usage limits follow your Cache key instead of your IP, giving you consistent access across machines, networks, and deployments.
No. Cache doesn't list, index, or provide search over prompts and answers. It returns a saved answer only when another request matches. Requests are otherwise handled with the same data considerations as sending them to the upstream AI provider. Pro encrypted cache additionally protects generated response content at rest.
Start by reusing cached answers