Moonshot AI releases Kimi K2.7-Code open-source model

Moonshot AI has released Kimi K2.7-Code, an open-source, coding-focused agentic model. Built on the K2.6 architecture, the model features 1 trillion total parameters, 32 billion activated parameters, and a 256K context window using a Mixture-of-Experts (MoE) structure. Benchmarks indicate significant improvements over its predecessor, particularly in the MLS Bench Lite, where it showed a 31.5% increase in performance, placing it competitive with frontier models like GPT-5.5.

As OpenAI continues to evolve its Codex platform toward full cloud-based operations with GPT-6 integration, domestic users in certain regions face persistent barriers including network connectivity, payment hurdles, and data compliance regulations. In response to these limitations, Kimi Work has emerged as a primary desktop alternative. Kimi Work provides a localized execution loop through its Goal mode, multi-agent collaboration, and specialized database connectivity, designed to handle tasks within domestic regulatory and technical environments.