Hi @andr11v,

What you are seeing appears to be two separate issues rather than a single root cause.

For the pay-per-token issue, the regional support table shows that Claude models are supported in AWS us-west-2, and the docs also say those models are available through pay-per-token endpoints in your workspace. In practice, though, regional support does not always mean every workspace in that region already has every Databricks-hosted Claude endpoint exposed. The fact that databricks-gpt-oss-20b works but databricks-claude-haiku-4-5 returns ENDPOINT_NOT_FOUND strongly suggests this is a workspace-level availability or enablement issue for the Claude endpoints, not a general Model Serving issue.

For Provisioned Throughput, that is a separate path from pay-per-token. The error message Provisioned throughput is not enabled for this workspace usually means the workspace or account does not have the required Databricks-side entitlement enabled, and that is not a self-service workspace toggle. Separately, for Databricks-hosted Claude models from system.ai, Claude is not currently supported for Provisioned Throughput, so that specific deployment path is not expected to work today.

From what I can gather, 

• Missing Claude pay-per-token endpoints is likely a workspace-level enablement or availability issue for Databricks-hosted Claude endpoints.
• Provisioned Throughput for Claude is not currently expected for Databricks-hosted Claude models from system.ai.
• The PT entitlement error itself is likely a separate workspace/account-level gating issue that Databricks Support or the account team would need to verify.

I would recommend opening a support case and asking Databricks to verify or enable Databricks-hosted Claude pay-per-token availability for this workspace, and also confirm whether the workspace has Provisioned Throughput entitlement enabled.

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