Kimi K2.7 Code
Kimi K2.7 Code is a coding-specialized mixture-of-experts model with 1 trillion total parameters and 32 billion active per token — 384 experts across 61 layers, MLA attention with a SwiGLU feed-forward path, and a MoonViT vision encoder for image and video input. It ships with native INT4 quantization and a mandatory thinking mode, reporting a 21.8% gain on Kimi Code Bench v2 over K2.6 while consuming 30% fewer reasoning tokens — built for agentic coding across large repositories.
Features
On-demand Deployments
On-demand deployments let you run Kimi K2.7 Code on dedicated GPUs with Sciforium's high-performance serving stack, with high reliability and no rate limits.
DocsToken-Efficient Agentic Coding
Uses 30% fewer reasoning tokens than K2.6 at higher accuracy, with native INT4 weights keeping large-repository work affordable.
Docs