Haiku 5.5’s lowest rate ends above 100K prompt tokens.

Haiku 5.5 is available as claude-haiku-5-5. Prompt length selects the price tier. Up to 100K tokens, Anthropic lists $0.10 input and $0.50 output per million tokens; above 100K, the rates are $0.50 and $2.50. Cache reads are $0.01/$0.05 and five-minute cache writes $0.125/$0.625 across the same threshold. These launch rates remain listed when checked on 19 October. ¹

Try bounded discovery, classification or summarization with a result you can check. Record each request’s prompt size and include the lead model’s review and retries in task cost. A short evaluation prompt can substantially understate the cost of a long agent conversation.

Anthropic also halves Sonnet 5.5’s cache-read price to $0.10 per million tokens. Its reported average savings depend on the workload’s cache share. Keep the context threshold visible in your evaluation rather than applying one discount to every agent turn. ²