Kimi AI: In addition, this model features a completely new model architecture, increasing the intelligence level per unit of computing resources by 2.5 times, rather than simply increasing the number of parameters. At the same time, besides Kimi K3, we will also open up more of the technology stack supporting the model's operation, including a high-performance attention computing kernel, the MoE communication library, and the infrastructure supporting large-scale intelligent agent environments.
