Kimi releases and open-sources Kimi-K2.7-Code

Kimi 发布并开源最新代码模型 Kimi-K2.7-Code

Kimi open-sourced K2.7-Code, scoring 11%–31.5% higher than K2.6 on three in-house benchmarks. Inference token usage dropped 30%, and long-coding-task instruction-following and end-to-end success rate both improved. A 6x speed mode is coming; the model is available now via Kimi API and Kimi Code. The post doesn't disclose parameter count, training data, or the open-source license.

Why it matters: Moonshot open-sourced a code model with solid gains on three in-house benchmarks and a 30% inference efficiency improvement — a real cost signal. No external benchmarks (LiveCodeBench, SWE-bench) or parameter count disclosed, so capped below 85. Still, a major Chinese lab open...