Apparently this course used half cascade models like gemini-2.0-flash-live-001, so tool calling worked well. However, now all these models have been fully replaced by native audio models like gemini-2.5-flash-native-audio-preview-09-2025.
The native audio models work for this course up to lesson 1, but for lesson 3 (tools for your agent) they don’t work well. See something here for a reference: Bring back the Half Cascade gemini-live-2.5-flash-preview model - Gemini API - Google AI Developers Forum
It seems that this type of native audio models are not mature enough for tool calling use cases. I would prefer to have half cascade models back.