> ## Documentation Index
> Fetch the complete documentation index at: https://docs.braintrust.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Bedrock model variations for on-demand throughput

> [!NOTE]
>
> **Applies to:**
>
> * Plan - Any
> * Deployment - Any
> *
> * Use case - AWS Bedrock integration with cross-region inference profiles

## Summary

AWS Bedrock model invocation fails with a `ValidationException` indicating that `on-demand throughput` is not supported when using base model identifiers like `anthropic.claude-sonnet-4` in the Braintrust playground. This occurs because base Bedrock model identifiers do not support on-demand throughput, which is required for cross-region inference profiles. The issue is resolved by selecting the model variation with the full version suffix (e.g., `global.anthropic.claude-sonnet-4-5-20250929-v1:0`) instead of the base model identifier.

## Resolution Steps

### Update model selection in playground

#### Step 1: Locate model variation

In the Braintrust playground model dropdown, find the model variation with the full version identifier including `global` prefix and version suffix.

#### Step 2: Select versioned model

Choose the variation format (e.g., `global.anthropic.claude-sonnet-4-5-20250929-v1:0`) rather than the base model name.

### Verify IAM permissions

#### Step 1: Confirm inference profile access

Ensure your IAM policy includes inference profile resources in addition to foundation models.

```text theme={"theme":{"light":"github-light","dark":"github-dark-dimmed"}}
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:InvokeModelWithResponseStream"
      ],
      "Effect": "Allow",
      "Resource": [
        "arn:aws:bedrock:*::foundation-model/*",
        "arn:aws:bedrock:*:*:inference-profile/*"
      ]
    }
  ]
}

```


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.