thought_signature Error 400 for gemini-3-flash

[markine]

Bug Description

Hi team. We're using gemini-3-flash with Livekit inference and Livekit cloud, running python agent 1.4.1.

Seeing thought_signature related errors of the type:

   12:34:42.038 WARNI… livekit.agents     failed to generate LLM completion, retrying in 0.1s                                                                                                            
                                           Traceback (most recent call last):                                                                                                                             
                                             File "/Users/XXXXXXXXXXX/.venv/lib/python3.13/site-packages/livekit/agents/llm/llm.py", line 206, in _main_task         
                                               return await self._run()                                                                                                                                   
                                                      ^^^^^^^^^^^^^^^^^                                                                                                                                   
                                             File "/Users/XXXXXXXXXXX/.venv/lib/python3.13/site-packages/livekit/agents/inference/llm.py", line 346, in _run         
                                               raise APIStatusError(                                                                                                                                      
                                               ...<5 lines>...                                                                                                                                            
                                               ) from None                                                                                                                                                
                                           livekit.agents._exceptions.APIStatusError: provider: google model: gemini-3-flash-preview, message: Error 400, Message: 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:YYYYYYYYYYYYY` , position 28. Please refer to https://ai.google.dev/gemini-api/docs/thought-signatures for    
                                           more details., Status: INVALID_ARGUMENT, Details: [].  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:YYYYYYYYYYYYY` , position    
                                           28. Please refer to https://ai.google.dev/gemini-api/docs/thought-signatures for more details. (status_code=400, request_id=None, body=provider: google        
                                           model: gemini-3-flash-preview, message: Error 400, Message: 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:YYYYYYYYYYY` ,        
                                           position 28. Please refer to https://ai.google.dev/gemini-api/docs/thought-signatures for more details., Status: INVALID_ARGUMENT, Details: [].  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:YYYYYYYYYYY` , position 28. Please refer to                                              
                                           https://ai.google.dev/gemini-api/docs/thought-signatures for more details., retryable=True)                        

Looking at recent logs 1.4.x promised to resolve this but no luck. Any advice? I'd be happy to help debug further, I just don't have the context.

Thanks!

Expected Behavior

No errors for gemini 3 flash after upgrading to 1.4.x.

Reproduction Steps

1. 
2.
3.
...
- Sample code snippet, or a GitHub Gist link -

Operating System

Linux

Models Used

gemini-3-flash

Package Versions

"livekit-agents[turn-detector]==1.4.1",
    "livekit-plugins-deepgram==1.4.1",
    "livekit-plugins-openai==1.4.1",
    "livekit-plugins-silero==1.4.1",
    "livekit-agents[google]==1.4.1",
    "livekit-plugins-google==1.4.1",

Session/Room/Call IDs

No response

Proposed Solution

Additional Context

No response

Screenshots and Recordings

No response

[adrian-cowham]

Relevant slack thread: https://livekit-users.slack.com/archives/C09J2AYGN0M/p1767906437874559

[adrian-cowham]

@markine Ok to close?

[markine]

@adrian-cowham Adrian, thank you for your collaboration! Resolved!

[aalkhulaifi605]

@markine @adrian-cowham currently facing the same issue, can you elaborate on how to fix it? i am on 1.4.1

[adrian-cowham]

This was addressed in the server-side LiveKit Inference code. Can you confirm you're using LiveKit Inference? If so, can you share the exception you're seeing?