Databricks expands Unity Gateway spend visibility, with billing and timing limits
Databricks’ October 27 release notes describe Unity Gateway Usage Dashboard v1.0, with expanded model and user spend analysis, tag-based attribution and coding-agent session analysis. Its October 26 entry separately announces the Spend Status API for checking usage against configured budgets. These are related visibility features, not proof that all model spending is free, real-time or automatically capped.
The usage dashboard draws on system-table data covering models, MCP services and coding agents. Current documentation says account administrators create an account- or workspace-scoped dashboard and select whether viewers use shared data permissions or their own. Access to the underlying tables still matters; creating a dashboard is not the same as making every user’s data visible to everyone.
Databricks separates request-level usage analytics from billing-based cost attribution. The usage table includes tokens, latency, requester details, request tags and MCP or coding-agent metadata. Records are typically available within an hour, but new workspaces can take longer, and scheduled dashboard refreshes run every six hours. Those delays matter when monitoring fast-moving workloads.
Usage tracking is a billed Unity Gateway feature. Databricks-hosted model cost attribution uses MODEL_SERVING records in the billing system table, while supported external model-provider services use token-based dollar estimates derived from published provider prices and configured price multipliers. The external estimates may differ from the final provider invoice and are for information, not guaranteed charges.
Tags can help attribute costs to teams or workloads, and the Coding Agents tab can identify sessions with high estimated spend. External spend tracking is limited to model-provider services in Unity Catalog and does not support the Custom provider. The October 26 Spend Status API entry offers budget-status retrieval; that announcement alone does not establish automatic request blocking.
The release notes warn that updates are staged and an account might receive a feature a week or more after its initial release. Current dashboard templates can also replace customizations when updated, so Databricks advises saving a copy first. Teams evaluating the dashboard should verify rollout, permissions, data latency and billing coverage before treating it as a complete live cost ledger.