Deploying Kimi K3 on Amazon SageMaker HyperPod and Amazon EKS

On July 27 2026 Moonshot AI released Kimi K3, a 2.8‑trillion‑parameter Mixture of Experts model that is the first open‑weight system in the 3‑trillion‑parameter class. The architecture—using Kimi Delta Attention, Gated Multi‑Head Latent Attention, and Stable LatentMoE—spreads weights over 896 experts but activates only 16 per token (≈104 B parameters per pass), enabling efficient long‑horizon coding, agentic workflows, and multimodal reasoning. The accompanying deployment guide demonstrates how to host Kimi K3 on AWS with either Amazon SageMaker HyperPod or a custom Amazon EKS cluster, leveraging high‑end GPU compute and optimized serving frameworks for large‑scale model inference.

Source: AWS ML Blog | 2026-07-30