# Kimi K2.7-Code: an open model that actually shipped its weights

> Moonshot's Kimi K2.7-Code (12 June 2026) is open-source with weights live on Hugging Face.

*After a month of 'open-weight, weights pending', a refreshingly literal release.*

By The InsidersFeed Desk · InsidersFeed
Canonical: https://insidersfeed.com/news/kimi-k2-7-code-actually-open

> **Key:** **The take:** after a season of labs calling things 'open-weight' before releasing weights, Moonshot did the radical thing — put the actual files on Hugging Face on day one. Low bar; still worth applauding when it's cleared.

**Kimi K2.7-Code** landed 12 June as a coding-specialised successor to April's K2.6. Modified MIT licence, weights live, commercial use fine. Moonshot claims ~21.8% over K2.6 on its in-house code benchmark and ~30% fewer reasoning tokens — i.e. cheaper to run for similar work.

## The honest size of it

Don't oversell it: this is an *incremental, vendor-benchmarked* coding model, not a frontier event, and 'better on our own benchmark' is the oldest move in the book. But it's open, it's real, and it's the kind of fast, free iteration that keeps grinding down the advantage of closed labs.

> **Note:** **The skeptic's footnote:** an in-house benchmark (Kimi Code Bench v2) is convenient marketing. Wait for LiveBench-style independent runs before you treat the 21.8% as gospel. The licence and the download, though, are facts you can check today.

## FAQ

### Is Kimi K2.7-Code genuinely open?
Yes — weights are on Hugging Face under a Modified MIT licence as of 12 June 2026, so you can download, run and use it commercially. That's a stronger claim than some recent 'open-weight' launches that shipped without the weights.

### Should I trust the 21.8% improvement?
Treat it as a vendor claim — it's measured on Moonshot's own Kimi Code Bench v2. The direction is plausible and the model is real; wait for independent benchmarks to pin down the actual gain.