Is there a way to tell/extract during or post call if Vapi uses a fallback model for the LLM/ASR/TTS models for a self defined assistant? Or even just tell which models were specifically evoked for a call?

Yes, Vapi lets you check which ASR, LLM, or TTS models were used during a call via detailed call logs or the GET /logs API, though explicit fallback indicators may not always be shown directly.

Seems that the GET /logs API may be deprecated based on the Vapi documentation (https://docs.vapi.ai/api-reference/logs/get). From the GET /call (https://docs.vapi.ai/api-reference/calls/get) it appears that the main distinction in terms of which model is used would potentially be contained within the costs component of the response, as the assistant component holds all of the model and fallback information without necessarily specifying the one used. However, the API for the costs component seems to only potentially match the base models for voice/model/transcriber and not fallback. Do you know if the fallback would therefore be reported here if it was used, or will it continue to default to the primary set models?

The GET /logs API is deprecated, and while the assistant field in GET /call shows configured models and fallbacks, only the costs field reflects the actual model used, so if a fallback was triggered, its usage should appear there.