400 {'error': {'message': 'Invalid JSON payload received. Unknown name "arguments" at \'tools[0].function_declarations[0]\': Cannot find field.)',

Gemini API
, ,
1

my error:
openai.BadRequestError: Error code: 400 - {‘error’: {‘message’: ‘Invalid JSON payload received. Unknown name “arguments” at ‘tools[0].function_declarations[0]’: Cannot find field. (request id: 2025122307592348564637tkUFmagZk)’, ‘type’: ‘’, ‘param’: ‘’, ‘code’: 400}}

my tools:

 {

                "type": "function",

                "function": {

                    "name": "AskUser",

                    "description": "主动向用户提问以获取更多信息或确认操作。",

                    "parameters": {

                        "type": "object",

                        "properties": {

                            "question": {"type": "string", "description": "要询问用户的问题"},

                            "options": {"type": "array", "description": "可选的回答选项", "items": {"type": "string"}}},

                        "required": ["question"]},

                            },

            },

I don’t encounter errors when calling the GPT model using the OpenAI method, but switching to the gemini-3-pro-preview model results in errors

all code:

def chat_async_tool_select(messages):

    response = client.chat.completions.create(

        model="gemini-3-pro-preview",

        messages=messages,

        tools=[

            {

                "type": "function",

                "function": {

                    "name": "AskUser",

                    "description": "主动向用户提问以获取更多信息或确认操作。",

                    "parameters": {

                        "type": "object",

                        "properties": {

                            "question": {"type": "string", "description": "要询问用户的问题"},

                            "options": {"type": "array", "description": "可选的回答选项", "items": {"type": "string"}}},

                        "required": ["question"]},

                            },

            },

        ],

       tool_choice="auto"

    )

    tool_use = response.choices[0].message.tool_calls

    response_content = response.choices[0].message.content

    output_tokens = response.usage.completion_tokens

    input_tokens = response.usage.prompt_tokens

    return tool_use, response_content, input_tokens, output_tokens

message:

messages = [{ 'role': 'system','content': 'you are a helpful assistant'},

              {'role': 'user', 'content': 'hello, i want get some informations'}]
5

Hi @beiluo, welcome to the community!

Gemini 3 models enforce much stricter validation than the 1.5 or 2.5 series, specifically regarding Thought Signatures .

These signatures are encrypted parts of the model’s response that represent its internal thinking. For Gemini 3, it is now mandatory to include the exact thought_signature from the model’s functionCall when you send the functionResponse back in the next turn. If that signature is missing or altered, the API returns a 400 error because the reasoning chain is broken.

Ensure your chat history logic preserves the entire Part object (including the signature) rather than just extracting the function arguments.

You can check out the official Gemini 3 Thought Signatures Guide for the exact JSON structure required.

Thank you!

7

It doesnt matter, even if you send the thought signatures with a totally inmuted Part