Coding-focused 1T MoE (32B active) with 256K context. +21.8% over K2.6 on Kimi Code Bench v2.
Agentic coding, long-horizon tasks, multi-step engineering
Purpose-built for autonomous coding agents
30% less thinking tokens than K2.6 (more efficient)
+21.8% Kimi Code Bench v2, +31.5% MLS Bench Lite
Strong instruction following for long-horizon tasks
Choose Kimi K2.7 Code for autonomous coding workflows where the model needs to plan, execute, and iterate over many steps. It's optimized for multi-file engineering tasks and reduces unnecessary reasoning overhead.
256K context is large but not 1M. For tasks needing broader context, M3 or Gemini may be better. Coding-focused — less versatile for general conversation.
42 models available. Switch per message. Start free.
Get Started Free