Haiku 5.5 Is So Cheap I Stopped Hoarding Tokens
At ten cents per million input, Claude's small model changed how I route work.
Anthropic released Claude Haiku 5.5 on October 7 at roughly 75% lower cost than its predecessor -- about $0.10 in and $0.50 out per million tokens. That number quietly changed my architecture.
Before, I'd lump every sub-task onto one big model to avoid the overhead of thinking about routing. Now I send classification, routing and short summaries to Haiku and keep the expensive model for the few steps that need judgment.
The saving isn't just the bill. It's that I can run the small model constantly, on every message, without flinching. The big model became a treat, not the default.
If you're building anything agentic, price per call matters more than peak score. Haiku 5.5 is the argument for designing around cheap routing.