Claude Haiku 5.5 Becomes Viable Sub-Agent: Human-Level Computer Use at Ninety Percent Lower Cost

Anthropic on Tuesday released Claude Haiku 5.5, and the figures it published do something that prior small-model launches have not managed: the model scored 72.4% on OSWorld 2.1, the standard benchmark for AI computer use — matching the 72% human baseline established in the original evaluation paper. It did so at $0.10 per million tokens for the vast majority of real-world requests, a 90% reduction from the $1.00 per million charged for its predecessor.

First reported by Tech Times. This page is a short excerpt; the full text is theirs.