Kimi K2.7 Code

About this model

Moonshot Kimi K2.7 Code is a 1‑trillion‑parameter Mixture‑of‑Experts (MoE) chat model with 32 B activated parameters, featuring a 256 K token context window and a MoonViT vision encoder (400 M parameters). Optimized for coding and agentic tasks, it operates in a forced thinking mode that preserves reasoning across turns. The model achieves 62.0 on the Kimi Code Bench v2, 53.6 on Program Bench, and 76.0 on the MCP Atlas benchmark. It supports text, image, and video inputs and provides function calling capabilities. The model is released under a custom “Other” license.

Technical specifications

Capabilities

Input modalities

Output modalities

Reasoning

Default on

Knowledge horizon