AWS Bedrock + Sonnet 5 / Opus 5: streaming responses silently dropped after reasoning block (works fine on direct Claude config) #1687
Replies: 4 comments 1 reply
|
Thanks for asking your question. Please be sure to reply with as much detail as possible so the community can assist you efficiently. |
|
Hey there! It looks like you haven't connected your GitHub account to your Deepgram account. You can do this at https://community.deepgram.com - being verified through this process will allow our team to help you in a much more streamlined fashion. |
|
It looks like we're missing some important information to help debug your issue. Would you mind providing us with the following details in a reply?
|
|
Hello @liz-at-dkg, Thank you for the detailed report and the request ID. Apologies for the delay in a reply. Your description matches a known problem with reasoning models on the I cannot find a documented setting that disables reasoning for
To help the team, please share the exact Bedrock model IDs, the region, and your |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Is anyone also experiencing this? I'm seeing an issue with Deepgram Voice Agent using Claude Sonnet 5 / Opus 5 through AWS Bedrock.
Setup
Issue
The first turn works normally. On later turns, once Bedrock returns a reasoningContent block through ConverseStream, Deepgram seems to silently drop the assistant response. No error is raised and the agent just continues listening as if no response was received.
Bedrock CloudWatch logs confirm the full response is returned correctly:
reasoningContent → text → stopReason: end_turnHowever, on Deepgram's next request to Bedrock, that assistant turn is completely missing from the message history. The new STT transcript is instead appended to the previous user turn.
The same models work through the direct Anthropic API, where reasoning is returned as a type: thinking block.
It looks like there may be an issue handling Bedrock's streamed reasoningContent events. Has anyone else seen this with Sonnet 5 / Opus 5 + Bedrock? Or a way to disable thinking from Deepgram for AWS Bedrock?
All reactions