Databricks Adds Moonshot AI’s Kimi K3
Databricks has added Moonshot AI’s Kimi K3 to its enterprise AI platform. The integration gives customers access to the open-weight model through the Databricks Foundation Model API. Moreover, Databricks provides governance through Unity AI Gateway and comprehensive zero data retention controls.
Kimi K3 is Moonshot AI’s latest open-weight model. It contains 2.8 trillion parameters and supports a context window of up to one million tokens. Furthermore, the model supports coding, agentic reasoning, and document understanding.
Kimi K3 brings open weights to enterprises
Databricks says Kimi K3 performs at the level of leading proprietary models on enterprise tasks. The company also cites a score of 57 on Artificial Analysis’ Intelligence Index. Consequently, businesses can consider the model alongside offerings from Anthropic, OpenAI, and Google.
The integration also focuses on enterprise data access. Through Unity Catalog, Kimi K3 can work with governed enterprise data, feature stores, and lakehouse information. Therefore, organizations can ground AI applications in their own business data while maintaining existing governance policies.
Governance, cost controls, and model choice
Unity AI Gateway provides a centralized control layer for Kimi K3 and other models. In addition, administrators can manage permissions, safety guardrails, audit trails, and spending limits. Databricks says these controls apply consistently across open-weight and proprietary models.
Cost is another focus of the integration. According to Databricks, Kimi K3 delivers 50% to 72% lower cost per task than comparable proprietary models. As a result, continuous coding agents and large document-processing workloads could become more economical at scale.
Kimi K3 is currently hosted by Databricks in the United States. Initially, the model supports AWS and Google Cloud workspaces, while Azure Databricks customers can access it through ADI. Meanwhile, Databricks plans to add more hosting regions.
The move also strengthens Databricks’ multi-model strategy. Customers can switch between Kimi K3, Anthropic Claude, OpenAI GPT, and Google Gemini through a unified API. Thus, enterprises can select models based on performance, cost, and specific workloads without rebuilding their applications.
