We researched a lot and find that for now it's openai's models as they have the best reasoning capabilities.

we plan to integrate different models for different use cases, i.e. google palm for easy questions like top 5 restaurants in the city but that require more up-to-date information