Haiku 5.5 is out, and the headline is the price: $0.10 per million input tokens and $0.50 per million output under 100K, roughly 75% cheaper to run on average than Haiku 4.5. It's available now across the Claude platform and on AWS, Google Cloud, and Azure, under the model ID `claude-haiku-5-5`. Anthropic pitches it at high-volume, repetitive work — summaries, classification, context compaction, database queries — and as a subagent under Opus 5.5 or Sonnet 5.5 on coding tasks. The new trick is an adjustable effort setting, a first for the Haiku line, with five levels from Low to Max so you can dial cost against intelligence per call. On Terminal-Bench 4.0 it scores 39.2% — well short of Sonnet 5.5's 70.6%, but a different planet from Haiku 4.5's 0.0%. One catch before you redo the budget: Simon Willison clocked the new tokenizer eating about 1.25x the tokens of 4.5 on the same prompt, and the company concedes the model uses slightly more tokens per task, so the real-world saving is smaller than the sticker cut. Claude Code already made it the default Haiku in the same-day 2.1.293 release.