I am seeing two related issues with Databricks-hosted Claude models in my workspace.
Workspace details:
- Cloud: AWS
- Region: us-west-2
- Tier: Premium PAYG
- Serverless compute: enabled
- Unity Catalog/metastore: configured
- Compliance security profile: disabled
- "Enforce data processing within workspace Geography for Designated Services": disabled
Issue 1: Claude pay-per-token endpoint is missing
According to the supported foundation models table, us-west-2 supports Databricks-hosted Claude models such as databricks-claude-haiku-4-5 and databricks-claude-opus-4-7 for Foundation Model APIs pay-per-
token.
However, when I call the chat completions endpoint with a Claude model, I get ENDPOINT_NOT_FOUND.
For example:
model="databricks-claude-haiku-4-5" -> ENDPOINT_NOT_FOUND
Pay-per-token inference works for databricks-gpt-oss-20b, so Model Serving itself appears to be working.
Issue 2: Provisioned Throughput is not enabled
Separately, when I try to create a model serving endpoint with provisioned throughput for a Databricks-hosted Claude model from system.ai in Unity Catalog, endpoint creation fails with:
"Provisioned throughput is not enabled for this workspace."
My questions:
1. Why are Databricks-hosted Claude pay-per-token endpoints missing from my workspace even though us-west-2 is listed as supported?
2. Is Provisioned Throughput for Claude models expected to be available in this workspace, or does it require a separate workspace/account-level entitlement?
