I didn’t realize they were tuning cost optimization by switching models contextually. That’s very clever. I bet the whole industry of consumer LLM apps moves that way.