By Asif Razzaq
Publication Date: 2026-06-13 04:57:00
This week Moonshot AI was released Kimi K2.7 code. It is a coding-oriented, agent-based model. The model weights are delivered on Hugging Face under a modified MIT license. You can also reach it via Kimi API and Kimi Code.
K2.7 code is aimed at long-term software engineering, not general chat. It plans, edits, runs tools and debugs in many steps. Moonshot combines the model with a subscription coding platform.
Kimi K2.7 code
K2.7 code is a mixture of experts model. It contains a total of 1T parameters and enables 32B per token. The design uses 384 experts, with 8 selected per token and 1 shared. It consists of 61 layers, including a dense layer.
Attention uses MLA and the feed forward path uses SwiGLU. A MoonViT Vision encoder adds 400 million parameters for image and video input. The model ships with native INT4 quantization. The context window contains 256,000 tokens (262,144).
Two restrictions are important: the mode of thinking is mandatory; Disabling it will return an API error. Sampling…
