Breaking

Moonshot Releases Coding-Focused Kimi-K2.7-Code as Open Weights

June 12, 2026 at 22:41 EDT

Software Dev & Coding
Foundation Models
Open Source

On June 12, 2026, Moonshot AI released the weights and code for Kimi-K2.7-Code, a coding-focused open-weight model, on Hugging Face under a Modified MIT License. Built on Kimi-K2.6, it is an agentic model optimized for long-horizon software engineering tasks involving tool use.

June 12 · Moonshot AI

Kimi-K2.7-Code: an open-weight coding model closing the gap on closed rivals

A 1.1-trillion-parameter Mixture-of-Experts model — built for long-horizon agentic software engineering, released with open weights under a Modified MIT License, and priced aggressively.

1.1T

Total parameters (MoE)

32B

Active at inference

256K

Context tokens (262,144)

−30%

Reasoning tokens used

Benchmark gains vs K2.6

Improvement on Moonshot's internal coding suites

+21.8%

Kimi Code
Bench v2

+11.0%

Program
Bench

+31.5%

MLS Bench Lite
(Py/Rust/Go)

API pricing

Per million tokens — positioned aggressively against closed models

$0.19

Cache hit

$0.95

Input

$4.00

Output

What developers praise

Physics-sim & HTML5 Canvas output on par with top closed models, at far lower cost
Strong agentic workflow with Kimi Code CLI
Long-horizon maintenance of complex codebases; end-to-end fixes

Caveats to watch

Benchmarks are internal; limited comparison to SWE-Bench Verified
Fewer reasoning tokens may mean shallower reasoning at times
Trillion-scale MoE is heavy to run locally (quantization advised)

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.