Claude Haiku 5.5
- Source
- anthropic.com
- Author
- sfkgtbor
- Date
Anthropic positions Haiku 5.5 as its cheapest and fastest model and the first Haiku with an adjustable effort setting, aimed at subagents, computer use and high-volume work. The same release halved Sonnet 5.5 cache reads to $0.10 per million tokens. Distribution was immediate: Cursor, GitHub Copilot and Devin all added the model on launch day. Cognition reports 58.4% on FrontierCode 1.1 at roughly an eighth of Sonnet 5's cost per task, and GitHub says it matched Sonnet 5 on many coding tasks with fewer steps. The caveats matter for budgeting. Simon Willison notes the tokenizer produces about 1.25 times as many tokens for the same text, which hides part of the cut, and Artificial Analysis measured roughly three times GPT-6 Luna's output tokens at max effort, with cost-per-task figures that do not yet reflect the long-prompt price step. Route short, parallel subagent calls to it, and measure real cost per task before moving long-context jobs.

Context
Anthropic's own page for Claude Haiku 5.5, its small, low-cost , says the model costs around 75 percent less to run than Haiku 4.5 on average. The page lists $0.10 per million input and $0.50 per million output tokens for requests up to 100,000 tokens and $0.50 and $2.50 beyond that, and notes an updated tokenizer that uses slightly more tokens per task.
Anthropic calls it its first Haiku-class model with an adjustable effort setting, positions it for high-volume, cost-sensitive and narrowly scoped tasks, and says Sonnet 5.5 and Opus 5.5 remain better for complex agentic coding. Its tables compare Haiku 5.5 with Haiku 4.5, GPT-6 Luna and Sonnet 5.5, but these are the company's own reported results. For third-party scoring, see Artificial Analysis's evaluation.
- Model — A computational system that uses learned patterns to produce an output from information supplied to it.
- Token — A piece of text, such as a word, part of a word or a punctuation mark, that a language model represents as numbers and handles as a unit.
- Benchmark — A fixed set of tasks with a stated way of scoring them, used to compare models or setups.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
Checking sign-in…
Loading comments…

