Kimi K2.6 released with long-horizon execution, agent swarms, and full-stack AI coding

Moonshot AI has released its latest flagship model, Kimi K2.6, introducing advanced coding features, long-horizon execution, and a collaborative agent swarm system. This release delivers strong gains in coding tasks that require persistence and generalization, with reliable performance across Rust, Go, and Python, and tasks from front-end to DevOps and performance tuning. In its internal Kimi Code Bench testing suite, Kimi K2.6 shows marked improvements over its predecessor, Kimi K2.5.

Building on this coding proficiency, Kimi K2.6 can convert simple text prompts into complete front-end interfaces, generating layouts with deliberate design, interactive elements, and rich animations such as scroll-triggered effects. The model is also adept at using image and video generation tools, enabling the creation of cohesive visual assets and refined hero sections. In addition, Kimi K2.6 expands into simple full-stack workflows, automating common processes from authentication and user interaction to lightweight database operations like session management and transaction logging.

The Agent Swarm upgrade brings a qualitative leap, autonomously coordinating multiple specialized agents for tasks including search, document analysis, deep research, and parallel multi-format content generation. This compositional system can deliver documents, websites, slides, and spreadsheets in a single run. These technologies power 24/7 autonomous agents like OpenClaw and Hermes, with ongoing improvements to reliability, API interpretation, and long-running safety. Kimi K2.6 is now available through Kimi.com, the Kimi app, the API, and Kimi Code.