On July 27, 2026, Moonshot AI released Kimi K3, a 2.8 trillion parameter Mixture of Experts (MoE) model, described as the first open-weight system to reach the 3 trillion parameter class. The model distributes its parameters across 896 specialist experts, activating only 16 per token, meaning approximately 104 billion parameters are active during any single forward pass, yielding a 2.5x improvement in scaling efficiency over its predecessor, Kimi K2. Kimi K3 features a context window of 1 Million Tokens and native multimodal capabilities (Text + Vision). It excels at long-horizon coding tasks, agentic workflows, and complex reasoning, supporting native tool calling, structured output, and an always-on thinking mode. The open weights are available on Hugging Face under the identifier moonshotai/Kimi-K3, distributed in MXFP4 format. Serving requires a vLLM day-0 inference container, with commits currently in vllm/vllm-openai:kimi-k3. Deployment requires a p6-b300 instance (ml.p6-b300.48xlarge), providing 8 NVIDIA B300 Blackwell Ultra GPUs. AWS offers two primary mechanisms to procure this capacity: Amazon SageMaker HyperPod and an Amazon Elastic Kubernetes Service (Amazon EKS) cluster.