Anthropic released the model on October 7, 2026. Prompts up to 100,000 tokens are billed at $0.10 per million input tokens and $0.50 per million output tokens, while longer requests cost $0.50 and $2.50. Anthropic puts average savings at 75%, saying 90% of past Haiku requests fit in the cheaper tier. This is also the first Haiku with effort controls, so cost can be weighed against intelligence one task at a time. On the OSWorld 2.1 desktop benchmark it scored 72.4%, up from 15.7% for Haiku 4.5. It is offered through the Claude Platform, AWS Bedrock, Google Cloud Vertex AI and Microsoft Azure. The intended setup has a larger model such as Fable or Opus doing the planning while Haiku handles the subtasks. Anthropic also halved Sonnet 5.5 cache read prices on the same day, which it expects to cut most agentic workloads by about 20%. For founders, cheap subagent tokens make wide multi-agent designs much less expensive. A capable small model at this price also puts more pressure on the margins of startups that resell inference.