Python: Bug: BedrockChatCompletion does not support Bedrock Application Inference Profiles

[denny-nolan-kochcc]

Describe the bug
When attempting to invoke a BedrockChatCompletion using an application inference profile, the model_id parameter is failing to resolve with the error Model ID {model_id} does not contain a valid model provider name..

This is because the Bedrock model provider resolver relies on the model_id parameter value containing the name of the model provider (ie 'anthropic', not to be confused with higher-level Bedrock model provider):

semantic-kernel/python/semantic_kernel/connectors/ai/bedrock/services/model_provider/bedrock_model_provider.py

Line 41 in 7d72f89

return next(provider for provider in cls if provider.value in model_id)

This model provider name is being used to map what parameters are being sent in the bedrock invoke_model() call to align the parameters with the given model provider.

When hosting more than one AI application in an AWS account, Application Inference Profiles are used so that model usage can be tracked by application (docs), rather than just by total usage of a model across the account. The code above seems to be oriented around the second approach, which uses a foundation model id or arn that contains the model provider name like this: us.anthropic.claude-sonnet-4-20250514-v1:0

Unfortunately, when hosting the same model in an application inference profile the value does not contain the model provider as you can see here: arn:aws:bedrock:us-east-1:<AccountID>:application-inference-profile/51i7czk0cv6z.

This feature is very important to productionizing and scaling the use of semantic kernel in our applications utilizing bedrock resources, and is a full blocker in our dev platform to using semantic kernel with Bedrock-hosted models as we require the use of application inference profiles for cost tracking.

To Reproduce
Steps to reproduce the behavior:

1. Create an application inference profile in Bedrock. I used us.anthropic.claude-sonnet-4-20250514-v1:0 as the modelSource but this could likely be any model hosted in Bedrock.
2. Set up a BedrockChatCompletion like this and provide the application inference profile arn from step 1 in the model_id parameter
3. attempt to invoke the BedrockChatCompletion
4. You will receive the error Model ID arn:aws:bedrock:us-east-1:<AccountID>:application-inference-profile/51i7czk0cv6z does not contain a valid model provider name.

Expected behavior
We would be able to provide an application inference profile arn along with a model provider id (ie anthropic) to the BedrockChatCompletion and successfully call the models defined in the inference profile.

Platform

• Language: Python
• Source: main branch
• AI model: Bedrock - any model using application inference profile
• IDE: VS Code
• OS: Mac/linux

[TaoChenOSU]

Related to #10941

[TaoChenOSU]

Hi @denny-nolan-kochcc,

Thank you for your report!

I am able to reproduce the issue.

The problem is that we need the model provider to map the request body to Bedrock and the response body to SK data structure. Since the application inference profile doesn't contain the model provider and it seems like there is no way to include that in the generated profile arn, we need to add an optional parameter that will allow callers to specify the provider.

[denny-nolan-kochcc]

@TaoChenOSU Thank you for looking into this so quickly! I'd really like to start putting out some agent accelerators in our dev platform using semantic kernel so a fix to this would be very valuable.

From my cursory review of the code I agree that adding an optional parameter for specifying the provider would likely be the smoothest path forward to support application inference profiles without introducing a breaking change.