Moonshot releases open-source Kimi K2.7-Code with lower token usage

Moonshot AI has open-sourced Kimi K2.7-Code, a coding model with 1 trillion total parameters (32B active), a 256K context window and 384 MoE experts. The company claims 30% lower reasoning-token usage than K2.6 while benchmark scores improved.

• 1T total parameters, 32B active, 256K context, 384 MoE experts
• Reasoning-token usage cut by 30% versus K2.6
• MLS Bench Lite rose from 26.7 to 35.1 (+31.5%)
• Weights on Hugging Face under Modified MIT License, API at platform.moonshot.ai