https://blog.bytebytego.com/p/token-spend-out-of-control-the-case?triedRedirect=true

Routing is a great method of reducing token spend. Using a society of task-specific small language models is another. Both work to ensure the 80% light repeatable tasks don’t wastefully end up with fat models and only the genuine 20% do. Different techniques, different dynamics, different tradeoffs.