Unexpected Behavior Using Claude Sonnet 4.6 with Tool Use

Support
1

Hi everyone,

Could you please review the “Conduct Research” step from the following run log?

https://app.mindstudio.ai/share/debugger/0cd5178d-5920-4aa5-952c-a296f47d14fa/916308b0-f2aa-4df2-826f-7af954dae02e

This step is a Generate Text block with Tool Use enabled.

It is working quite well with Google models (using the Google Search tool). However, when I tested the same setup with Claude Sonnet 4.6, I encountered several issues:

  • Extremely long processing time

  • Significantly higher cost (compared to Google models using the Google Search tool, where the cost is only a small fraction of this)

  • No meaningful or usable output

  • The run log indicates that the code_execution tool was used, even though I did not enable it

I tried using Clause Sonnet 4.6 as an alternative in the Hackathon Workspace, so the cost impact is not critical in this case. However, I would like to better understand what might be going wrong in this configuration so I can avoid similar issues in the future.

Is there something specific about how Claude handles tool use in this context that I may be missing?

Thanks in advance for your insights.