
Haiku 5.5 Triage: Measure Effort, Then Route by Cost
Summary
Benchmark Haiku 5.5 effort levels on your own tickets, then route cheap-first.
Anthropic shipped Claude Haiku 5.5 today, October 7, 2026, and the headline is the price: $0.10 per million input tokens and $0.50 per million output tokens for requests under 100K tokens, down from $1 and $5 on Haiku 4.5. It is also the first Haiku with the effort setting, defaulting to medium, and the first with five levels to choose from.
A cheaper model does not automatically make your bill smaller. Effort level, thinking mode, a new tokenizer and a pricing cliff at 100K prompt tokens all move the number. So instead of guessing, this guide builds a small harness that runs your own labelled tickets through several Haiku 5.5 configurations, reports accuracy, latency and cost for each, picks the cheapest one that clears your quality bar, and then wraps it in a cheap-first router that escalates only when the answer looks wrong.
Keep reading — it's free
Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.
Already a member? Sign in
Comments
Be the first to comment