Missing thought_signature error using Gemini 3.0-flash-preview

[janghyun0402]

Error

[2026-01-21 15:58:23,091: ERROR/ForkPoolWorker-1] Error processing message: 400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'Unable to submit request because function call `diagnose_responsibility_tool` in the 10. content block is missing a `thought_signature`. Learn more: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/thought-signatures', 'status': 'INVALID_ARGUMENT'}}
[2026-01-21 15:58:23,094: ERROR/ForkPoolWorker-1] Traceback (most recent call last):
  File "/home/kindredpm/workspace/R2R_PM/AA_ADK_maintenance.py", line 4086, in process_message
    async for event in self.channel.agent_runner.run_async(
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/runners.py", line 505, in run_async
    async for event in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/runners.py", line 493, in _run_with_trace
    async for event in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/runners.py", line 722, in _exec_with_plugin
    async for event in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/runners.py", line 482, in execute
    async for event in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/agents/base_agent.py", line 294, in run_async
    async for event in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/agents/llm_agent.py", line 460, in _run_async_impl
    async for event in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/flows/llm_flows/base_llm_flow.py", line 370, in run_async
    async for event in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/flows/llm_flows/base_llm_flow.py", line 447, in _run_one_step_async
    async for llm_response in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/flows/llm_flows/base_llm_flow.py", line 816, in _call_llm_async
    async for event in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/flows/llm_flows/base_llm_flow.py", line 800, in _call_llm_with_tracing
    async for llm_response in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/flows/llm_flows/base_llm_flow.py", line 1053, in _run_and_handle_error
    raise model_error
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/flows/llm_flows/base_llm_flow.py", line 1039, in _run_and_handle_error
    async for response in agen:
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/models/google_llm.py", line 262, in generate_content_async
    raise ce
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/adk/models/google_llm.py", line 241, in generate_content_async
    response = await self.api_client.aio.models.generate_content(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/genai/models.py", line 7018, in generate_content
    return await self._generate_content(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/genai/models.py", line 5836, in _generate_content
    response = await self._api_client.async_request(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/genai/_api_client.py", line 1432, in async_request
    result = await self._async_request(
             ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/genai/_api_client.py", line 1365, in _async_request
    return await self._async_retry(  # type: ignore[no-any-return]
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/tenacity/asyncio/__init__.py", line 111, in __call__
    do = await self.iter(retry_state=retry_state)
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/tenacity/asyncio/__init__.py", line 153, in iter
    result = await action(retry_state)
             ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/tenacity/_utils.py", line 99, in inner
    return call(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/tenacity/__init__.py", line 420, in exc_check
    raise retry_exc.reraise()
          ^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/tenacity/__init__.py", line 187, in reraise
    raise self.last_attempt.result()
          ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/concurrent/futures/_base.py", line 449, in result
    return self.__get_result()
           ^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/concurrent/futures/_base.py", line 401, in __get_result
    raise self._exception
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/tenacity/asyncio/__init__.py", line 114, in __call__
    result = await fn(*args, **kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/genai/_api_client.py", line 1310, in _async_request_once
    await errors.APIError.raise_for_async_response(response)
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/genai/errors.py", line 203, in raise_for_async_response
    await cls.raise_error_async(status_code, response_json, response)
  File "/home/kindredpm/miniconda3/envs/BE/lib/python3.12/site-packages/google/genai/errors.py", line 225, in raise_error_async
    raise ClientError(status_code, response_json, response)
google.genai.errors.ClientError: 400 INVALID_ARGUMENT. {'error': {'code': 400, 'message': 'Unable to submit request because function call `diagnose_responsibility_tool` in the 10. content block is missing a `thought_signature`. Learn more: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/thought-signatures', 'status': 'INVALID_ARGUMENT'}}

Environment

• Google ADK 1.22.0
• Python 3.12.3
• Gemini 3.0-flash-preview + thinking enabled
• Ubuntu 24.04

Our Code

Call run_async

            async for event in self.channel.agent_runner.run_async(
                user_id=self.channel.user_identifier,
                session_id=self.channel.session_id,
                new_message=user_content
            ):
                if event.content and event.content.parts:
                    for part in event.content.parts:
                        if hasattr(part, "thought") and part.thought:
                            self.planning_text += part.text
                        elif event.is_final_response() and hasattr(part, 'text'):
                            response_text += part.text

                        await self.channel.initialize_adk_components(
                            agent_name="KindredLeasing",
                            agent_description=f"AI leasing agent via {channel_name}",
                            model=model,
                            system_prompt=leasing_prompt,
                            tools=all_tools,
                            before_tool_callback=agent_before_tool_callback,
                            after_tool_callback=agent_after_tool_callback,
                            plugins=[tool_logger_plugin],
                            planner=BuiltInPlanner(
                                thinking_config=thinking_config
                            ) if channel_name != 'voice' else None
                        )

Generate Agent and Runner

def initialize_adk_components():

            agent_kwargs = {
                'name': agent_name,
                'description': agent_description,
                'model': model,
                'instruction': system_prompt,
                'tools': tools
            }

            # Add callbacks if provided
            if before_tool_callback:
                agent_kwargs['before_tool_callback'] = before_tool_callback
            if after_tool_callback:
                agent_kwargs['after_tool_callback'] = after_tool_callback

            # Add any additional kwargs (like planner config)
            agent_kwargs.update(kwargs)

            self.agent = agents.Agent(**agent_kwargs)

            runner_kwargs = {
                'app_name': app_name,
                'agent': self.agent,
                'session_service': self.session_service,
                'memory_service': self.memory_service
            }

            # Add plugins if provided
            if plugins:
                runner_kwargs['plugins'] = plugins

            self.agent_runner = runners.Runner(**runner_kwargs)

After migrating LLM model from Gemini 2.5-pro to Gemini 3.0-flash-preview, in some cases when multiple tool call happens, missing thought_signature error occurs

I would like to know that it is resolved in 1.23.0 (latest version), or we should resolve this error in our code

[adk-bot]

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

TLDR: Yes, upgrading to google-adk version 1.23.0 or newer should resolve this error. The issue is that the new Gemini model you are using requires a thought_signature for certain tool calls, and support for this was explicitly added in version 1.23.0.

---

Hello! I can help with this.

The 400 INVALID_ARGUMENT error you're seeing comes from the Gemini API. Newer models, especially when thinking is enabled, require a thought_signature to be included with function calls. Your current version of google-adk (1.22.0) does not appear to have full support for this requirement with the model you're using.

The good news is that this has been addressed in a more recent version of the library.

Based on the official changelog, version 1.23.0 introduced a feature to "Allow thinking_config in generate_content_config". [1] This change was specifically made to handle requirements like the thought_signature.

Therefore, you should resolve this error by upgrading your google-adk package to the latest version:

pip install --upgrade google-adk

This will ensure that the framework correctly handles the API requirements for the newer Gemini models.

---

[1] See the ADK Changelog.