Suppose a support team receives hundreds of short messages each day. Before a person replies, an AI could label each one “billing,” “delivery,” or “technical help.” That repeated, narrow task is the sort of work Anthropic says its new Claude Haiku 5.5 is designed to handle.
What is Haiku 5.5?
Haiku is the small, speed-focused member of Anthropic's Claude model family. The October 7 release targets quick summaries, classification, database questions, live support, and simpler steps within a larger AI workflow. It is Anthropic's first Haiku model with an adjustable effort setting: a developer can favor a cheaper, quicker answer or spend more computation on a harder step.
Anthropic says Haiku 5.5 costs about 75% less per task than Haiku 4.5 on average. That is the company's comparison, not a promise for every job. API pricing also changes for requests above 100,000 input tokens, so a long document can have a different cost from a short message. The model is available through Claude and major cloud platforms; developers can select claude-haiku-5-5 in the Claude API.
When would you choose a bigger model?
A small model can handle the first pass on routine work, while a person or a stronger model reviews difficult cases. For example, Haiku might sort the support messages; Opus or Sonnet might investigate a complex bug or plan a longer coding task. Anthropic says Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding. Its October 7 update also cut Sonnet 5.5's cache-read price in half, which can lower costs for repeated workflows.
The practical test is your own task: check a sample of real messages for mistakes, measure response time, then decide which jobs need a stronger model or human review. Anthropic's benchmark charts are useful evidence for its claims, but they do not tell you the accuracy of your specific workflow.