Moonshot AI's Kimi K3 Model Launches on Amazon Bedrock with Improved Efficiency

Moonshot AI's Kimi K3 model has arrived on Amazon Bedrock, marking a significant milestone in the development of open-weight models. As described by Moonshot AI, Kimi K3 is its most capable model to date, boasting a 1-million-token context window and native vision capabilities. This combination results in an approximate 2.5x improvement in scaling efficiency over Kimi K2.

The model's architecture has undergone two updates: Kimi Delta Attention and Attention Residuals. These changes aim to improve information flow across sequence length and model depth. Additionally, Kimi K3 employs a mixture-of-experts design called Stable LatentMoE, which activates 16 of 896 experts.

Moonshot AI acknowledges that Kimi K3 still trails proprietary models like Claude Fable 5 and GPT 5.6 Sol in overall performance. However, it has delivered frontier-level results across its evaluation suite and consistently outperformed other tested models. The company also lists limitations, including the model's reliance on preserved thinking history mode, which can lead to unstable generation quality if an agent fails to pass back historical thinking content.

The launch of Kimi K3 on Amazon Bedrock reflects sustained investment in open-weight models by AWS. Since 2025, Bedrock has added dozens of open-weight models from various providers, including Moonshot AI. Customers can access Kimi K3 through the US Geo and Global cross-Region inference profiles, with the full list of supported Regions available in the Bedrock documentation.