Kimi K3 Model Now Available on Databricks via Unity AI Gateway

Kimi K3, Moonshot AI's advanced open-weight model, is now available on Databricks via the Foundation Model API. It supports a 1 million token context window and native multimodal capabilities, delivering performance comparable to proprietary models.

Kimi K3 is integrated with Databricks' Unity AI Gateway, providing enterprise-grade governance, including access permissions, safety guardrails, audit trails, and spend controls. This ensures consistent policy application across all models on the platform.

The model is designed to offer frontier-level quality for enterprise tasks such as coding, agentic reasoning, and document understanding at a fraction of the cost of comparable proprietary models, often 50-70% lower per task.

On Databricks, Kimi K3 can directly access and reason over governed enterprise data stored in the lakehouse and feature stores through Unity Catalog, grounding responses in an organization's specific knowledge.

Databricks provides customers with the flexibility to use Kimi K3 alongside other open-weight and proprietary models, enabling intelligent routing and easy swapping of models without application code changes.

Kimi K3 is available now for US-hosting on Databricks Foundation Model API for AWS and GCP workspaces, with ADI access for Azure Databricks. Users can explore it in AI Playground, configure Unity AI Gateway, or build agents with Agent Bricks.