Summaries, classification and the smaller jobs delegated by coding agents are the workloads Anthropic is targeting. The model, identified as claude-haiku-5-5, uses adaptive thinking with adjustable effort. Platform notes specify a context window of 1 million tokens and a maximum output of 128,000 tokens. Anthropic's performance and savings claims describe its own evaluation, not an independent comparison.
The price depends on the size of the entire prompt. Up to 100,000 tokens, input costs $0.10 and output $0.50 per million tokens. Above that threshold, the corresponding rates are $0.50 and $2.50. This is not a progressive surcharge applied only to the tokens beyond 100,000, so the threshold matters when estimating the cost of a task.
Haiku 5.5 is available through Claude Platform and the major cloud channels, including AWS, Google Cloud and Azure; Claude Code supports it from v2.1.293. Belgian and EEA access still depends on cloud regions and contractual terms. Existing integrations are not necessarily drop-in compatible: older thinking, sampling and assistant-prefill parameters can cause 400 errors, making the migration guide relevant even when the task itself is unchanged.
The official launch date is October 7. An email received on October 8 does not move it into a later day's announcements, and the Opus 5.5 and Sonnet 5.5 releases remain September events. Haiku is the new launch here, with a position centered on volume, latency and cost rather than a fresh launch of the whole model family.
