Instead there will be choice, and people will go for the cheapest model that is “good enough” for the task at hand. Do you need the most cutting edge expensive model to proof read an email?

That’s already happening now, with companies telling their employees to be mindful of their usage and use lower reasoning levels for mundane/basic stuff (documents, summaries, etc.) and reserve the more costly models and reasoning for the extensive tasks (i.e. coding). The AI companies are now starting to charge money and suddenly tokenmaxxing isn’t such a good idea for management.