benchrCache miss / 1M$3input
Cache hit / 1M$0.30input
Output / 1M$15hosted API
Context1.05M131K default output
Official price table
| Tier | Input / 1M | Output / 1M | Cached input / 1M |
|---|---|---|---|
| Cache miss | $3 | $15 | — |
| Cache hit | $0.30 | $15 | $0.30 input |
Cache reuse drives the hosted decision
A repeated long prefix can make input ten times cheaper on a cache hit, while output remains $15. Measure hit rate and output volume together. The open weights create a separate deployment option, but hardware, expert routing, observability, and availability are not represented by the hosted token price.
Frequently asked questions
Does Kimi K3 have long-context surcharges?
Moonshot describes flat pay-as-you-go rates with no context-length tier.
What is a cache hit priced at?
$0.30 per million input tokens, compared with $3 on a cache miss.
Is self-hosting free?
No. The weights are published, but infrastructure and operations remain your cost.