Kimik2.7 Code is a coding-focused, open-source agentic model developed by Moonshot AI, designed to plan, edit, run tools, and debug across many steps rather than produce single-shot answers. It is built as a coding-first system with forced thinking and preserved reasoning traces, so the model always deliberates before responding rather than offering an instant mode. The model targets long-horizon software engineering work, including multi-file refactors, repository-scale context, and end-to-end tasks, while generalizing across mainstream languages used in backend, infrastructure, performance, systems, security, frontend, and ML/data engineering.

Relative to its predecessor K2.6, Kimik2.7 Code improves coding capability and agentic task execution while using roughly thirty percent fewer reasoning tokens. It posts large gains on coding benchmarks such as Kimi Code Bench v2, Program Bench, and MLS Bench Lite, and it also advances agentic performance on autonomous task suites including Kimi Claw 24/7 Bench, MCP Atlas, and MCP Mark Verified, with the latter reaching the low eighties and outperforming Claude Opus 4.8 in that comparison. Native image and video input alongside text output, combined with first-class MCP-style tool use, make it a practical fit for teams that need an open-weight coding agent capable of sustaining coherent, tool-driven workflows across long sessions.