[BUG] INVALID_ARGUMENT using gemini-3-pro-preview

[dpolistwm]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

1.17.0

Python Version

3.12

Operating System

Ubuntu 24.04.1

Installation Method

pip

Steps to Reproduce

Simply changing the model_id in GeminiModel to gemini-3-pro-preview from gemini-2.5-flash causes the exception after a tool call:

File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/_api_client.py", line 1225, in _async_request_once
await errors.APIError.raise_for_async_response(response)
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/errors.py", line 188, in raise_for_async_response
await cls.raise_error_async(status_code, response_json, response)
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/errors.py", line 210, in raise_error_async
raise ClientError(status_code, response_json, response)
google.genai.errors.ClientError: 400 Bad Request. {'message': '{\n "error": {\n "code": 400,\n "message": "Request contains an invalid argument.",\n "status": "INVALID_ARGUMENT"\n }\n}\n', 'status': 'Bad Request'}

Expected Behavior

Successful execution

Actual Behavior

INVALID_ARGUMENT exception

Additional Context

No response

Possible Solution

No response

Related Issues

No response

[afarntrog]

Hi @dpolistwm, thanks for raising this! I can reproduce it with the below.

from strands import Agent, tool
from strands.models.gemini import GeminiModel
import os
@tool
def letter_counter(word: str, letter: str) -> int:
    """
    Count occurrences of a specific letter in a word.

    Args:
        word (str): The input word to search in
        letter (str): The specific letter to count

    Returns:
        int: The number of occurrences of the letter in the word
    """
    if not isinstance(word, str) or not isinstance(letter, str):
        return 0

    if len(letter) != 1:
        raise ValueError("The 'letter' parameter must be a single character")

    return word.lower().count(letter.lower())


model = GeminiModel(
    client_args={"api_key": os.getenv('GOOGLE_API_KEY')},
    model_id="gemini-3-pro-preview",
)
agent = Agent(model=model, tools=[letter_counter])

result = agent("how many letters are in 'hello' use the letter_counter tool ")

Error:

Tool #1: letter_counter

Tool #2: letter_counter

Tool #3: letter_counter

Tool #4: letter_counter
cycle failed
......

This seems to be related to the following updates (from the docs)
https://ai.google.dev/gemini-api/docs/thought-signatures

As a general rule, if you receive a thought signature in a model response, you should pass it back exactly as received when sending the conversation history in the next turn. When using Gemini 3 Pro, you must pass back thought signatures during function calling, otherwise you will get a validation error (4xx status code).

[Ratish1]

Hello @afarntrog , I'd like to contribute to this.

[afarntrog]

Great @Ratish1 ! Please see this https://github.com/strands-agents/sdk-python/blob/main/CONTRIBUTING.md and feel free to get started.

[Ratish1]

I have read the contributing guidelines @afarntrog and have followed it, and I have created the PR for this issue.

[dpolist]

I have downloaded the GeminiModel implementation directly from the specific PR commit (https://github.com/strands-agents/sdk-python/blob/32fe567e25cd72ad9c7d638d4107e789392cfa66/src/strands/models/gemini.py), made quick adjustments to the imports:

from strands.types.content import ContentBlock, Messages
from strands.types.exceptions import ContextWindowOverflowException, ModelThrottledException
from strands.types.streaming import StreamEvent
from strands.types.tools import ToolChoice, ToolSpec
from strands.models.gemini import validate_config_keys
from strands.models.model import Model

And the problem persists (maybe for different reasons, though):

Using latest prompt version: 29
Successfully started func1_mcp_client
Successfully started func2_mcp_client
Successfully started func3_mcp_client

Tool #1: itemDetails
Setting attribute on ended span.
Tried calling set_status on an ended span.
Tried calling _add_event on an ended span.
Calling end() on an ended span.
cycle failed
Traceback (most recent call last):
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/strands/event_loop/event_loop.py", line 189, in event_loop_cycle
async for tool_event in tool_events:
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/strands/event_loop/event_loop.py", line 531, in _handle_tool_execution
async for event in events:
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/strands/event_loop/event_loop.py", line 277, in recurse_event_loop
async for event in events:
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/strands/event_loop/event_loop.py", line 152, in event_loop_cycle
async for model_event in model_events:
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/strands/event_loop/event_loop.py", line 396, in _handle_model_execution
raise e
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/strands/event_loop/event_loop.py", line 337, in _handle_model_execution
async for event in stream_messages(
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/strands/event_loop/streaming.py", line 457, in stream_messages
async for event in process_stream(chunks, start_time):
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/strands/event_loop/streaming.py", line 391, in process_stream
async for chunk in chunks:
File "/home/danielp/r-d-pocs/strands-ai-trek/src/models/GeminiModel.py", line 437, in stream
raise error
File "/home/danielp/r-d-pocs/strands-ai-trek/src/models/GeminiModel.py", line 375, in stream
response = await client.models.generate_content_stream(**request)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/models.py", line 7185, in generate_content_stream
response = await self._generate_content_stream(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/models.py", line 5928, in _generate_content_stream
response_stream = await self._api_client.async_request_streamed(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/_api_client.py", line 1393, in async_request_streamed
response = await self._async_request(http_request=http_request, stream=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/_api_client.py", line 1309, in _async_request
return await self._async_retry( # type: ignore[no-any-return]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/tenacity/asyncio/init.py", line 111, in call
do = await self.iter(retry_state=retry_state)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/tenacity/asyncio/init.py", line 153, in iter
result = await action(retry_state)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/tenacity/_utils.py", line 99, in inner
return call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/tenacity/init.py", line 420, in exc_check
raise retry_exc.reraise()
^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/tenacity/init.py", line 187, in reraise
raise self.last_attempt.result()
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/concurrent/futures/_base.py", line 449, in result
return self.__get_result()
^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.12/concurrent/futures/_base.py", line 401, in __get_result
raise self._exception
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/tenacity/asyncio/init.py", line 114, in call
result = await fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/_api_client.py", line 1225, in _async_request_once
await errors.APIError.raise_for_async_response(response)
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/errors.py", line 188, in raise_for_async_response
await cls.raise_error_async(status_code, response_json, response)
File "/home/danielp/r-d-pocs/strands-ai-trek/.venv/lib/python3.12/site-packages/google/genai/errors.py", line 210, in raise_error_async
raise ClientError(status_code, response_json, response)
google.genai.errors.ClientError: 400 Bad Request. {'message': '{\n "error": {\n "code": 400,\n "message": "Request contains an invalid argument.",\n "status": "INVALID_ARGUMENT"\n }\n}\n', 'status': 'Bad Request'}

Works perfectly fine with gemini-2.5-flash

[Ratish1]

Hi @dpolistwm, I pushed a new commit. I think the thought signature wasnt extracted from the API response and sent back with the subsequent tool call history correctly. It should prevent the INVALID_ARGUMENT error now.

[dpolistwm]

Hey @Ratish1, almost there:

google.genai.errors.ClientError: 400 Bad Request. {'message': '{\n "error": {\n "code": 400,\n "message": "Unable to submit request because function call default_api:itemDetails in the 2. content block is missing a thought_signature. Learn more: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/thought-signatures",\n "status": "INVALID_ARGUMENT"\n }\n}\n', 'status': 'Bad Request'}

[dpolistwm]

Guys, it looks like other implementations are needed in strands-agents to preserver thoughSignature in the message history... I think I have a fix here... Can I have a go?

[Ratish1]

@dpolistwm , I'm working on it right now, I havent added thought signature to the tool result, that was the problem. I can push the commit soon, if you want to try it out. Also, wdym by other implementations?

[dpolistwm]

I had to change

src/strands/types/tools.py
src/strands/types/content.py
src/strands/event_loop/streaming.py

To allow for thoughtSignature preservation...

[dpolistwm]

I can add the diffs here ;) It's all now working fine

[Ratish1]

Oh ok so is my implemenation completely wrong @dpolistwm , or is yours a complementary to mine?.