# Moonshot ships Kimi K2.7-Code — open weights, coding-focused

> Moonshot released open-source Kimi K2.7-Code on 12 June 2026, weights on Hugging Face.

*Another open model from China, and this one's actually downloadable. The brief.*

By The FeaturedDaily Desk · FeaturedDaily
Canonical: https://featureddaily.com/news/kimi-k2-7-code-brief

> **Key:** **The one-liner:** Moonshot kept the Chinese open-weight conveyor belt running — a coding-focused Kimi update, weights on the table from day one.

**What happened.** On 12 June, Moonshot AI released **Kimi K2.7-Code**, an open-source successor to April's K2.6, with weights on Hugging Face under a Modified MIT licence. It's tuned for coding and autonomous code tasks.

**The claims.** Moonshot reports roughly 21.8% improvement over K2.6 on its own *Kimi Code Bench v2*, gains on Program Bench, and about 30% lower reasoning-token usage — all vendor-reported (attribute, don't assert).

> **Note:** **Why it matters.** Genuinely downloadable, capable coding models keep shipping from Chinese labs on a near-monthly cadence — the steady pressure on closed, pay-per-token incumbents. Independent benchmarks will confirm the size of the gain.

## Key takeaways

- What: Kimi K2.7-Code, a coding-specialised open-source model from Moonshot AI.
- Licence: Modified MIT, weights on Hugging Face — genuinely downloadable and commercial-use OK.
- Claims: ~21.8% over K2.6 on Moonshot's Kimi Code Bench v2; ~30% lower reasoning-token use (vendor).
- Context: successor to April's K2.6 — part of a fast Chinese open-weight cadence.

## FAQ

### What is Kimi K2.7-Code in one line?
An open-source, coding-focused AI model from China's Moonshot AI, released 12 June 2026 with weights on Hugging Face — a successor to April's Kimi K2.6.

### Are the gains independently verified?
Not yet — the ~21.8% figure is on Moonshot's own benchmark. The weights are public, though, so independent testing should follow quickly.