Haiku 5.5 at Ten Cents Made Me Route 80 Percent of My Calls to the Cheap Model

Claude Haiku 5.5 costs about ten cents per million output tokens, and that single number changed how I build. Most of the calls I make are high-volume and low-stakes: classification, summarization, routing. They do not need the flagship.

I used to send everything to the big model because it was easiest. Then the bill arrived and I learned that 'easiest' was quietly the most expensive. Haiku does the boring 80 percent for a fraction and I barely notice the quality drop.

The 20 percent that is hard, nuanced, or high-stakes still goes to the expensive model. That is the whole insight: match the model to the stakes, not to your habits.

If you are not routing by cost yet, you are overpaying for the privilege of not thinking about it. The cheap model is good enough far more often than the pricing page implies.