What changed
Anthropic introduced Claude Haiku 5.5 on 7 October, positioning it for high-volume tasks such as classification, summarisation and scoped subagent work. The release adds adjustable effort to the Haiku family.
Anthropic also reduced Sonnet 5.5 cache-read pricing. Its published savings estimates depend on workload and token usage; they are not a guaranteed reduction for every application.
Our perspective
For an AI application builder, this is a reason to evaluate routing by task. Try a smaller model on bounded steps with clear acceptance criteria, while retaining escalation for difficult or ambiguous work.
Compare the cost of a successfully completed task, including retries and review. Lower token prices are useful only when the workflow still meets its quality target. Keep a representative evaluation set before changing production defaults.
Go to the source
Primary reporting behind this brief. Our perspective is editorial analysis.
- Anthropic — Introducing Claude Haiku 5.5 7 Oct 2026