Skip to content

Mooncake

A KVCache centric serving platform for Kimi

Visit official site
Category AI models

What mooncake is

The platform is the Kimi serving system from Moonshot AI, built around a KVCache centric design. It separates prefill and decoding clusters to reuse idle resources.

What you can do with it

  • Balance throughput against latency SLOs
  • Apply early rejection under heavy overload
  • Serve long context requests

Getting started

  • Open the GitHub page
  • Read the technical report
  • Set up the environment and use the API

Pros & cons

✓ What we like

  • Long context scenarios show strong results
  • The material cites up to 525% throughput gains in some simulations

! What to watch out for

  • Environment setup follows the project docs
  • Integration uses the provided API or interfaces

FAQ

Who provides it?

Moonshot AI, as the serving platform for Kimi.

Is research published?

Yes, an open technical report is available to the community.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review