What changed
Anthropic introduced Claude Haiku 5.5 on October 7 for high-volume work including summarization, subagents and browser use. Routing remains an engineering decision.
Why it matters
A smaller model can make a good first pass when a task has a clear output contract. The dangerous case is silent failure: a plausible summary that omits an exception or a classification that triggers an unintended workflow. Route by measurable task requirements, not a vague division between easy and hard prompts.
What to check next
Define an output schema, a held-out test set and an escalation rule before changing routing. Measure how often the system needs a second attempt. Count that extra work in latency and cost, and keep permissions narrow regardless of model size.
Event dates refer to the original announcement. Publication dates show when this article first appeared on Codorc.
