I have a similar issue. The chat playground works fine until an AI Search data source is connected, then two or three messages in, the rate limit exceeded message shows.

The error message provides a link to the deployed model in Azure Open AI. Like you I have the tokens per minute set high.

As the limit message only appears when the AI search service is connected, it seems to suggest it either hitting a limit there, or using the connected service somehow increases the tokens/requests per minute consumed per prompt at the model.

The AI Search service is on the basic tier, and there's nothing to suggest (looking under monitoring) that any usage limits are being reached there either.

Any insights here would be welcome.