It clearly shows that tool calling was successful and provided a result. But the AI keeps gaslighting and telling me that the results haven't come in yet. What could be causing this? Call ID: 019ae3cb-7d41-7cc2-8b84-790b029312bb https://cdn.discordapp.com/attachments/1445727355753594931/1445727356193865728/image.png?ex=69433246&is=6941e0c6&hm=e1b84b3218421176590dca2816aabef2aff6d4ebd3302b7cd3a78f877135c668&

Thank you for sharing the details of your experience. Based on the information provided, it appears that the tool calling was successful and returned a result, but the AI did not acknowledge or use that result as expected. Unfortunately, I do not have access to specific technical logs or internal diagnostics that would allow me to determine the exact cause of this behavior. If you have any additional details or patterns you've noticed, feel free to share them, and I'll do my best to help with the information available.

Ok so switching to a different model seems to solve this issue immediately. Does GPT-Realtime Cluster not support tool calls?

+1 to this issue. I'm having the same problem

Your assistant prompt needs explicit instructions on how to handle scheduling appointments. You are correct that it has the data, but it just doesn't know what to do with it. Add instructions in the prompt like this example:

Once the `check_calendar_availability` tool has returned some results, ask what time the user would like to schedule. if they don't have a specific time in mind, suggest three of the times from the result. once the user has picked a time, use the `create_calendar_booking` and create a calendar event for that time.

It's still not reliable after adding those specific instructions. Here's a call with GPT Realtime Cluster: 019aed68-e8ae-7ff9-9850-52ebff81b86f I swapped model to GPT 5.1 without changing anything else and these were the results: 019aed6b-2255-7ccd-9bdb-104c2c92c94a

We typically recommend using non-realtime models for our platform as they perform better consistently

So I'm going to presume this means that you are not planning on fixing these issues anytime soon?

Our team has been making some updates to realtime implementations recently and there is no ETA for the fix specifically for tool calling with realtime.

Ok thanks for the update!