Gemini-3.0-pro / Gemini-3.0-flash failing to trigger tool_call

[GinkoJack911]

When my root agent uses gemini-3.0-pro or gemini-3.0-flash, it fails to initiate a tool_call. Is this because I’ve enabled the Planner(), or is there another reason? I’ve noticed that the model simply stops running after it finishes "thinking," rather than actually triggering the tool or calling a sub-agent.

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: This issue likely stems from a compatibility nuance between the newer Gemini 3.0 models and the ADK framework's handling of tool calls after the BuiltInPlanner completes its "thinking" phase. To diagnose this, the best first step is to run your agent with verbose=True to get more detailed output.

---

Hello! Thanks for reaching out with this detailed question.

The behavior you're describing, where the model "finishes thinking" and then stops, is characteristic of how the BuiltInPlanner works. [14] The model first generates a plan (the "thinking" part), and when it decides to use a tool, it outputs a special instruction. The ADK framework is then supposed to detect this instruction, execute the corresponding tool, and send the result back. [4, 22] The problem you're encountering seems to be a breakdown in that second step, where the framework isn't executing the tool after the plan is made.

This could be due to a few reasons, especially with newer models like the Gemini 3.0 series. Here are a few troubleshooting steps you can take to narrow down the cause:

1. Enable Verbose Logging: The most helpful step is to see what the ADK runner is doing under the hood. If you are using runner.run_debug(), you can enable this by setting the verbose flag. This will print detailed information about tool calls and results. [2]

   await runner.run_debug("Your query", verbose=True)

   Look for lines like [Calling tool: ...] in the output. If you don't see this, it confirms the framework is not attempting to execute the tool.

2. Update Your ADK Version: Support for new models and bug fixes for tool calling are added regularly. [1] Please ensure you are on the latest version of google-adk:

   pip install --upgrade google-adk

3. Test with a Different Model: To isolate whether the problem is with your agent's setup or the model itself, try temporarily switching to a more established model like gemini-2.5-flash. [8] If tool calls work correctly with that model, it strongly suggests the issue is specific to how the Gemini 3.0 models are interacting with the planner.

Could you please try these steps and share the output from the verbose log? That would provide more insight into what's going wrong.

[2] publicly accessible HTTPS URL of the document
[4] publicly accessible HTTPS URL of the document
[8] publicly accessible HTTPS URL of the document
[14] publicly accessible HTTPS URL of the document
[22] publicly accessible HTTPS URL of the document