[Bug]: Thought Signatures not returned with Responses API for Gemini 3 #17469
Description
What happened?
Hello,
When using the /responses api, the thought signature is not returned.
Is it plan to support the responses api for gemini3 as well ? I have been trying to use for tool calls, but without the tool signature, there isn't much we can do.
I have been using aresponses with gemini/gemini-3-pro-preview
Relevant log output
input to aresponses litellm
"model": "gemini/gemini-3-pro-preview",
"stream": false,
"num_retries": 5,
"drop_params": true,
"temperature": 1.0,
"reasoning_effort": "low",
"max_tokens": 4096,
"tools": [
{
"name": "get_horoscope",
"parameters": {
"type": "object",
"properties": {
"sign": {
"type": "string",
"description": "An astrological sign like Taurus or Aquarius"
}
},
"required": [
"sign"
]
},
"strict": null,
"type": "function",
"description": "Get today's horoscope for an astrological sign."
}
],
"input": [
{
"type": "message",
"content": "You are a helpful assistant. You first think about the reasoning process in the mind and then provide the user with the answer.",
"role": "system"
},
{
"type": "message",
"content": "What is my horoscope for today? My sign is Taurus.",
"role": "user"
}
]
}
Model dump output of the ResponsesAPIResponse object
{
"id": "resp_bGl0ZWxsbTpjdXN0b21fbGxtX3Byb3ZpZGVyOmdlbWluaTttb2RlbF9pZDpOb25lO3Jlc3BvbnNlX2lkOk1EOHhhZjdrSWJlRC04WVAycW50LVFN",
"created_at": 1764835114,
"error": null,
"incomplete_details": null,
"instructions": null,
"metadata": {},
"model": "gemini-3-pro-preview",
"object": "chat.completion",
"output": [
{
"type": "reasoning",
"id": "rs_7209472162028847676",
"status": "completed",
"role": "assistant",
"content": [
{
"type": "output_text",
"text": "**Taurus Horoscope - Let's Get This Done!**\n\nOkay, so the user wants their Taurus horoscope for today. Easy peasy. They've been crystal clear about their sign. And thankfully, I've got this nifty `get_horoscope` tool that's just perfect for the job. The prompt even practically wrote the code for me! They've already specified `sign='Taurus'`, so no ambiguity there. No need for any fussy refinement either, which is always a bonus. Alright, let's fire up that tool and get them the astrological insights they're after. Consider it done.\n\n\n",
"annotations": []
}
]
},
{
"type": "message",
"id": "MD8xaf7kIbeD-8YP2qnt-QM",
"status": "completed",
"role": "assistant",
"content": [
{
"type": "output_text",
"text": null,
"annotations": []
}
]
},
{
"arguments": "{\"sign\": \"Taurus\"}",
"call_id": "call_0e05ace67f524b7ca11d6e3da971",
"name": "get_horoscope",
"type": "function_call",
"id": "call_0e05ace67f524b7ca11d6e3da971",
"status": "completed"
}
],
"parallel_tool_calls": false,
"temperature": 0.0,
"tool_choice": "auto",
"tools": [],
"top_p": null,
"max_output_tokens": null,
"previous_response_id": null,
"reasoning": {},
"status": "completed",
"text": {},
"truncation": null,
"usage": {
"input_tokens": 179,
"input_tokens_details": null,
"output_tokens": 103,
"output_tokens_details": null,
"total_tokens": 282,
"cost": null
},
"user": null,
"store": null
}
Error raised after tool call output is being provided.
"Function call is missing a thought_signature in functionCall parts. This is required for tools to work correctly, and missing thought_signature may lead to degraded model performance. Additional data, function call `default_api:get_horoscope` , position 2. Please refer to https://ai.google.dev/gemini-api/docs/thought-signatures for more details."
Are you a ML Ops Team?
No
What LiteLLM version are you on ?
1.79.0
Twitter / LinkedIn details
No response