Microsoft has expanded its Microsoft Foundry platform with the addition of Moonshot AI’s Kimi K3 model, offered through Fireworks AI. Microsoft Azure announced the model is live on July 31, describing it as a 2.8-trillion-parameter open-weight model with a 1-million-token context window. Kimi K3 is a Mixture-of-Experts (MoE) model developed by Moonshot AI, placing it among the largest open-weight models currently available. Due to its sparse design, only a slice of the network activates for each token during inference, with the technical paper stating it activates 104 billion parameters. The build uses 896 routed experts, selecting 16 per token through the Stable LatentMoE setup. The 1-million-token context window allows the model to handle large software repositories, long documents, and extended agent workflows in a single run. Fireworks describes Kimi K3 as Moonshot AI’s most capable flagship model, confirming both the parameter scale and context window. The model is positioned for complex work including long-horizon coding, knowledge work, reasoning, vision, and agentic workloads. Its architecture includes Kimi Delta Attention, Attention Residuals, and Stable LatentMoE, with post-training involving reinforcement learning across general, coding, and agentic domains. This setup provides developers on Microsoft’s AI platform another large open-weight option for applications relying on long context, coding strength, deeper reasoning,