Fix: Tool message ordering for Gemini models

[habema]

Summary

Fixes Missing corresponding tool call for tool response message errors when using Gemini models. The message ordering fix that was already applied to Anthropic models is now also applied to Gemini models, since they seem have the same requirement that tool calls must immediately precede their corresponding tool responses.

Error:

litellm.exceptions.APIConnectionError: litellm.APIConnectionError: Missing corresponding tool call for tool response message. Received - message={'role': 'tool', 'tool_call_id': 'call_c1f7ac7f4d564f0ab55574de3052', 'content': '{"success": true, "result": {"webBuildErrors": {}}, "error": null}'}, last_message_with_tool_calls={'role': 'assistant', 'content': "It's currently 12:16 PM. Is there anything I can help you with in your app today?"}

Repro Script:

import asyncio
from datetime import datetime

from agents import Agent, Runner, function_tool

# function_call_output preceeds function_call
history_list = [
    {
        "type": "message",
        "role": "user",
        "content": [{"type": "input_text", "text": "Hi. What time is it?"}],
    },
    {
        "call_id": "call_04de935c15ad4e4b9ac226e5c2e8",
        "output": '{"success":true,"result":{"time":"12:16 PM","format":"%I:%M %p"},"error":null}',
        "type": "function_call_output",
    },
    {
        "arguments": '{"format": "%I:%M %p"}',
        "call_id": "call_04de935c15ad4e4b9ac226e5c2e8",
        "name": "get_server_time",
        "type": "function_call",
        "id": "__fake_id__",
        "provider_data": {
            "model": "vertex_ai/gemini-3-flash-preview",
            "response_id": "Cc9sadGmNLCAz_IPqPSEqQw",
            "thought_signature": "CiEBjz1rX65+CQhe4AGRK95x8byJ82UbcroiDLuw+NNisjI=",
        },
    },
    {
        "id": "__fake_id__",
        "content": [
            {
                "annotations": [],
                "text": "It's currently 12:16 PM. Is there anything I can help you with in your app today?",
                "type": "output_text",
                "logprobs": [],
            }
        ],
        "role": "assistant",
        "status": "completed",
        "type": "message",
        "provider_data": {
            "model": "vertex_ai/gemini-3-flash-preview",
            "response_id": "C89sab_sG-mEz_IPvYGN0AY",
        },
    },
]


@function_tool
def get_server_time() -> str:
    """
    Returns the current time.
    """
    return datetime.now().strftime("%I:%M %p")


agent = Agent(
    name="Assistant",
    model="litellm/vertex_ai/gemini-3-flash-preview",
    instructions="You are a helpful agent.",
    tools=[get_server_time],
)


async def main():
    result = await Runner.run(agent, history_list + [{"content": "How about now?", "role": "user"}])
    print(result.final_output)


if __name__ == "__main__":
    asyncio.run(main())