Kimi K2.7-Code: open-source coding model with better token efficiency

Kimi K2.7-Code is a new open-source coding model that improves upon its predecessor K2.6 with better performance on real-world software engineering tasks while reducing thinking-token usage by approximately 30%. The model uses a Mixture-of-Experts architecture with 1 trillion total parameters and demonstrates competitive performance across coding and agentic benchmarks, outperforming K2.6 on tasks like code generation, program reconstruction, and multi-day collaborative workflows.