Databricks Hosted Foundation Models usage and costs

[MA4]

1. Before proceeding with experimentation and subsequent production use of the Databricks‑hosted Foundation Models (e.g., Claude, Llama, Gemma via Foundation Model APIs), could you please confirm whether a separate agreement with the underlying model provider is required in addition to our existing Databricks Master Cloud Services Agreement?
2. Since we are using Azure Databricks, as a first-party service on Microsoft Azure cloud, Standard Databricks hosted LLMs pay per token usage would show up in our unified Azure bill, please confirm.
3. However, if we choose to use the external models hosted outside Databricks e.g. Azure Open AI we may need an agreement with that provider and billed separately.

[Ashwin_DSA]

Hi @MA4,

For Databricks-hosted foundation models that you access via Foundation Model APIs (for example, Claude, Llama, Gemma on Azure Databricks), you do not need to sign a separate commercial contract with Anthropic/Meta/Google in addition to your Databricks Master Cloud Services Agreement. These models are sold and operated as Databricks services. You do, however, need to comply with the model-specific terms and acceptable-use policies referenced in the Databricks docs (see Applicable model terms) in addition to your Databricks MCSA. You can get the list of supported models here. I'd recommend your legal team review those terms.

Your assumption is right. On Azure Databricks, usage of these Databricks-hosted FM APIs (pay-per-token or provisioned throughput) is metered as part of your Azure Databricks service and charged via your Azure bill, alongside your other Azure Databricks consumption.

If instead you configure external models such as Azure OpenAI via Mosaic AI Model Serving, you bring your own API key and commercial relationship with that provider. Those external model calls are billed by the external provider under your agreement with them. Databricks just provides the proxy, governance, and observability layer on top.

If you have an existing Databricks account team, they can also help you review your specific commercial setup and documentation links for your workspace and region.

Hope this helps.

If this answer resolves your question, could you mark it as “Accept as Solution”? That helps other users quickly find the correct fix.