The * gemini-2.5-flash-preview-native-audio-dialog model doesnt seem to make function calls at all even though the docs say that it supports it. And i can gaurantee that its not an implementation issue as as soon as i switch just the model to the live 001 function calling starts to work.
Furthermore, in some instances ive seen it to generate code in turn to execute it as a function call which is also wrong?
Can anyone help? Has anyone faced the same issue or has anyone actually gotten the model to make function calls consistently?
3 Likes
This used to work well, but in the past week has become increasingly hit or miss. My personal theory is they spin up a python vim behind the scenes and if it fails to allocate, no tool calling for you! I find that if you restart the session repeatedly until tool calling works, you’re then fine, but it’s incredibly frustrating and unreliable right now.
1 Like
Yea I can also vouch for how restarting the session many many times gets it to work rarely but other than that i guess its just completely broken for now.
This seems like a major issue, also see the ticket here - function calling is not working for gemini-2.5-flash-preview-native-audio-dialog · Issue #843 · googleapis/python-genai · GitHub
Very active issue many people are complaining about but have no heard any official response from google.
3 Likes
Any updates in regards to this?
Hi @Fawaz_Ali, @aljazdolenc @Israel_Cohen @Noah_Shipley1,
Thank you for reporting. I was unable to reproduce this issue on my end using AI Studio.
Can you please confirm if you’re still experiencing it?
Thank you!
Hey Guys,
Just now I have observed there is an inconsistency. Thank you for bringing this to our attention. We truly appreciate you flagging this issue and we have escalated it to our Internal team.
Thank you!
2 Likes
Hi @Fawaz_Ali,
In the meantime, could you please provide more information regarding your setup, such as your use case and the number of functions you have defined?
Thank you!
for me i used a a 1200 token prompt , with only one function to endCall , it works well with old model , but for flash 2.5 its not working at all
I can confirm that this happens to us as well, the function is called 1 out of 20 times for us, sometimes the model output audio of the function’s description instead. From what I’ve seen, this problem happens less on Google AIStudio for some reason but it’s prevalent when using the API. You can find a github issue of other people having the same problem mentioned above.
Hi @Ahmed_Mamdouh,
Thank you for your patience. We currently don’t have any updates to share at this moment. Rest assured, we will provide you with any new information as soon as it becomes available.
Thank you!
