They might just be early. The smaller models are getting more and more capable, for high-frequency use-cases it'll probably be worth using local quantized mode

They might just be early. The smaller models are getting more and more capable, for high-frequency use-cases it'll probably be worth using local quantized models vs paying for API inference.