Bedrock Application Inference Profile Support

Support Help
1

Does anyone know if Application Inference Profile’s are supported when routing AI requests through Bedrock? Fundamentally I’m looking to segment AWS bedrock costs by team. Being able to create two separate inference profiles for anthropic.claude-sonnet-4-5-20250929-v1:0 and having both options show up in cursor would be ideal.

I’m guessing this is not currently supported given that Cursor maintains the list of supported bedrock models and there’s probably no way to propagate the inference profiles arn to the model select window.

3

Hey, good question.

Application Inference Profiles aren’t supported yet. Cursor uses the standard Bedrock base model IDs, so you can’t pass a custom inference profile ARN through the model picker.

As a workaround, you can set up separate Cursor workspaces for each team, each with its own IAM role and Bedrock config, so you can track costs at the AWS role level. It’s not as clean as inference profiles, but it works.

4

Hey @deanrie, but you don’t see the Bedrock costs by IAM Role in AWS Cost Explorer? How does that work for you?

5

Hey @Hristo_Stoyanov good question. My workaround wasn’t quite accurate. Cost Explorer doesn’t filter Bedrock usage by IAM Role out of the box.

More correct options to split Bedrock costs by team:

  • Separate AWS accounts per team via AWS Organizations. This is the cleanest approach, and Cost Explorer will show costs per account.
  • Cost allocation tags. You can tag via STS session tags when you assume a role, then activate those tags in Billing for Cost Explorer.
  • CloudWatch or CloudTrail. Track usage per role and manually correlate it with Bedrock costs.

But these are all workarounds. Native support for Application Inference Profiles in Cursor would solve this much more cleanly.

6

Hey @deanrie, are there any plans for Cursor to support Application Inference Profiles soon?

7

Hey @Viktor_Marinov, right now there’s no specific timeline for supporting Application Inference Profiles. Cursor still uses the standard Bedrock base model IDs, and there’s no way to pass a custom inference profile ARN.

For now, the workarounds from my reply above separate AWS accounts and cost allocation tags are still the best options for splitting costs by team.