llama3-3-70b-instruct-v1:0 with on-demand throughput isn’t supported, but can not buy Provisioned Throughput

The error you're encountering is because Meta Llama 3 models in Amazon Bedrock require the use of an inference profile rather than direct on-demand throughput invocation.

To solve this problem, you need to:

1. Create or use an existing inference profile that includes the Meta Llama 3 model you want to use.
2. In your Python SDK code, use the ID or ARN of the inference profile instead of directly specifying the model ID.

This is a common requirement for certain models in Bedrock that don't support direct on-demand invocations. The Bedrock playground likely uses inference profiles behind the scenes, which is why it works there.

For Meta Llama models specifically, you'll need to use a cross-region inference profile. Amazon Bedrock provides system inference profiles for these models that you can use without having to create your own.

When working with the Python SDK, make sure to reference the inference profile ID/ARN rather than the direct model ID in your code. This approach is required for models like Meta Llama 3 that don't support direct on-demand invocation.
Sources
Flow Bedrock: Error Knowledge Bases | AWS re:Post
meta.llama3-2-11b-instruct-v1:0 not supported for on-demand throughput? | AWS re:Post
Choose teacher and student models for distillation - Amazon Bedrock