[BUG] Error 400 ""thinking.type.enabled" is not supported for this model" still exists in 2.1.113 when used with Opus 4.7 in Bedrock
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?
2.1.113 changelog claims to have fixed the error. However, using a Bedrock custom application profile based on Opus 4.7 still fails just like it did with 2.1.112:
hi
⎿ API Error: 400
{"type":"error","request_id":"req_txjtk5tatvmot5cmubv5hf3ulq2zsqyjhf3jwip2e5tt4lu6dsyq","error":{"type":"invalid_request_error","message":"\"thinking.type.enabled\"
is not supported for this model. Use \"thinking.type.adaptive\" and \"output_config.effort\" to control thinking behavior."}}
Tried this with and without ANTHROPIC_DEFAULT_OPUS_MODEL_SUPPORTED_CAPABILITIES=effort,xhigh_effort,max_effort,thinking,adaptive_thinking,interleaved_thinking
2.1.114 also has this problem.
What Should Happen?
Claude Code should function normally
Error Messages/Logs
Steps to Reproduce
- Configure an application inference profile in Bedrock based on the Opus 4.7
- Set AWS_PROFILE=x, CLAUDE_CODE_USE_BEDROCK=1, ANTHROPIC_MODEL='arn:...'
- Launch Claude Code, prompt it anything
Claude Model
Opus
Is this a regression?
No, this never worked
Last Working Version
_No response_
Claude Code Version
2.1.113
Platform
AWS Bedrock
Operating System
macOS
Terminal/Shell
iTerm2
Additional Information
_No response_
10 Comments
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
Claude Code 2.1.114 with bedrock encountered the same issue with application inference profile when switching to 1m context, while working well with 200K context. And system inference profile is working well too when switching to 1m context.
/model opus[1m]
⎿ Set model to arn:aws:bedrock:us-west-2:12345678901:application-inference-profile/abcdefg1234[1m]
❯ check the vercel sdk details.
⎿ API Error: 400 {"type":"error","request_id":"req_nqrrmqyswiqreseeqz4wmjfjx7xvhfi3scyzx7h2zclv24mmuohq","error":{"type":"invalid_request_error","message":"\"thinking.type.enabled\" is not supported for this
model. Use \"thinking.type.adaptive\" and \"output_config.effort\" to control thinking behavior."}}
Yes, I see the same: I didn't think to test the ARN without "[1m]", tried it now, and it does work with the smaller context. So it's the [1m] that breaks things.
Same problem with bedrock and claude v2.1.118 opus 4.7 with 1m context. Same model works fine with small default context and 4.6 work with 1m context.
(OS: Linux)
Can i confirm that this the https://github.com/anthropics/claude-code/issues/51711#issuecomment-4316854505 address this issue
It doesn't. A permission issue wouldn't cause the same application profile to behave differently with or without the [1m] suffix.
How ads you logging in? IAM rule or.using a token?
aws sso login, SSO through IAM Identity Center and a role assumption to the account with the app profile.
Preempting your next question - I have full admin rights in the account. Claude agrees:
Effective Bedrock perms: all bedrock:* actions (InvokeModel, InvokeModelWithResponseStream, ListFoundationModels, CreateModelCustomizationJob, etc.) on all resources inaccount xxxxxxxxx
thanks. this is a different issue than #51711 (that one was limited to token authentication). i think i can reproduce this locally now
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.