PERMISSION_DENIED: Endpoint databricks-claude-3-7-sonnet is not allowed

[mohnishbhagde]

I am working on a use case in my personal Databricks account - AgenticDataPipeline. 

I am getting below error when using  Anthropic Claude 3.7 Sonnet AI model

15:52:56.267 | ERROR | [Pipeline] Task failed: ACCOUNT DATA NORMALIZATION: ops_bronze.accounts_ra - 403 Client Error: PERMISSION_DENIED: Endpoint databricks-claude-3-7-sonnet is not allowed to be used by your workspace. Please reach out to Databricks to enable this endpoint for your workspace or upgrade your workspace tier. 

[szymon_dybczak]

Hi @mohnishbhagde ,

The access to some Anthropic models (like Claude 3.7 Sonnet) could be limited due to high demand. In some cases, these models won’t appear in the UI or you can encounter permission denied error like in your example. This was explain in below thread:

https://learn.microsoft.com/en-us/answers/questions/4373119/azure-databricks-claude-sonnet-endpoints...

[mohnishbhagde]

Thanks @szymon_dybczak  How can I get access or do I have to wait? Is there anything that I can do to finish up my use case?

[szymon_dybczak]

Hi @mohnishbhagde ,

Probably some waiting will do the trick. But maybe some databricks employee will jump in to discussion and shed some light.

[momo0101]

Hi @mohnishbhagde, 

has this error been resolved? If it has been resolved, how was it fixed? Is it a configuration issue?

(I have the same issue and contacted Databricks about a month ago and have been communicating with them, but they still don't seem to know the cause and it hasn't been fixed yet.)

[WiliamRosa]

Hi @mohnishbhagde,
As @szymon_dybczak mentioned, there have been some instabilities when using this endpoint. As an alternative, you could try another model such as Llama or Claude 4, or even configure an external model pointing directly to the Anthropic API. Also, keep in mind that a higher workspace tier is usually less prone to these issues (I’m currently using the Premium tier at work), although the additional cost might not be feasible for every scenario.