[BUG] Effort does not work with Bedrock when using Application Inference Profiles

Resolved 💬 5 comments Opened Feb 19, 2026 by brent-at-aam Closed Apr 27, 2026

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

There was a similar issue here: https://github.com/anthropics/claude-code/issues/23583

But I think the behavior persists when using Application Inference Profiles. Our profile is associated with Opus 4.6 but we don't see the option for effort:

Select model
Switch between Claude models. Applies to this session and future Claude Code sessions. For other/previous model names, specify with --model.  
                                                                                                                                                 
   1. Default (recommended)                                                                Use the default model (currently                      
                                                                                           arn:aws:bedrock:us-east-1:000000000000:application-in 
                                                                                           ference-profile/0aaaa0aaaa00)                         
   2. arn:aws:bedrock:us-east-1:000000000000:application-inference-profile/0aaaa0aaaa00    Custom Sonnet model                                   
 ❯ 3. arn:aws:bedrock:us-east-1:000000000000:application-inference-profile/0bbbb0bbbb00 ✔  Custom Opus model                                     
   4. arn:aws:bedrock:us-east-1:000000000000:application-inference-profile/0cccc0cccc00    Custom Haiku model                                    
                                                                                                                                                 
 ▌▌▌ Effort not supported for arn:aws:bedrock:us-east-1:000000000000:application-inference-profile/0bbbb0bbbb00

 Enter to confirm · Esc to exit

I'm assuming this is because there's no way to know by the ARN alone if it's Opus 4.6...

If we take a look at the profile we can see it's associated correctly:

aws bedrock get-inference-profile --inference-profile-identifier 0bbbb0bbbb00
{
    "inferenceProfileName": "my.us.anthropic.claude-opus-4-6-v1-claude-code",
    "description": "Cost allocation tracking profile for claude-code using us.anthropic.claude-opus-4-6-v1",
    "createdAt": "2026-02-06T00:55:38.963545+00:00",
    "updatedAt": "2026-02-06T00:55:38.963545+00:00",
    "inferenceProfileArn": "arn:aws:bedrock:us-east-1:000000000000:application-inference-profile/0bbbb0bbbb00",
    "models": [
        {
            "modelArn": "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-opus-4-6-v1"
        },
        {
            "modelArn": "arn:aws:bedrock:us-east-2::foundation-model/anthropic.claude-opus-4-6-v1"
        },
        {
            "modelArn": "arn:aws:bedrock:us-west-2::foundation-model/anthropic.claude-opus-4-6-v1"
        }
    ],
    "inferenceProfileId": "0bbbb0bbbb00",
    "status": "ACTIVE",
    "type": "APPLICATION"
}

Is there any way we can help claude to know it's opus 4.6?

Please let me know if I can provide any additional information to debug the issue.

What Should Happen?

claude knows the inference profile is mapped to opus 4.6 and supports effort.

Error Messages/Logs

Steps to Reproduce

  1. Create an application inference profile associated to Opus 4.6
  2. Setup claude to use the application inference profile arn as the default model for opus: ANTHROPIC_DEFAULT_OPUS_MODEL
  3. Launch claude and run /model

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

v2.1.47

Platform

AWS Bedrock

Operating System

Ubuntu/Debian Linux

Terminal/Shell

VS Code integrated terminal

Additional Information

_No response_

View original on GitHub ↗

This issue has 5 comments on GitHub. Read the full discussion on GitHub ↗