Anthropic released Claude Haiku 5.5 on Wednesday, the third launch in the Claude 5.5 family after Opus 5.5 and Sonnet 5.5 earlier this fall. Where the bigger siblings brought capability, Haiku 5.5 brings aggression: Anthropic calls it the cheapest, fastest, and most capable small model it has ever shipped, and the pricing backs up the claim.
For prompts of up to 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. Compared with Haiku 4.5's $1 and $5, that is roughly a 90 percent reduction. Prompts above the 100K-token threshold are priced at $0.50 and $2.50 respectively, still a 50 percent cut. Anthropic says the model costs around 75 percent less to run on average, noting that about 90 percent of Haiku 4.5 requests fell within the cheaper band.
The fine print deserves a read. A new tokenizer counts the same text as roughly 30 percent more tokens than Haiku 4.5's, which trims the headline savings in practice. And anyone who regularly pushes past 100K tokens will find the math changes above the threshold, where Anthropic's savings figure drops to 50 percent.
There is substance behind the price tag too. Haiku 5.5 is the first Haiku-class model with an adjustable effort setting, letting developers trade cost against intelligence on the same request rather than switching model tiers. It carries a one-million-token context window, up from 200K on Haiku 4.5, with up to 128K output tokens and a June 2026 knowledge cutoff. It is live on the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure, and free-tier Claude users can select it in the web, iOS, and Android apps.
The launch page carried two more announcements. Anthropic halved Sonnet 5.5 cache reads from $0.20 to $0.10 per million tokens, which it says makes Sonnet 5.5 around 20 percent cheaper on most agentic work. And paid plans now come with monthly API credits: $100 for Max 5x subscribers, $200 for Max 20x, and up to $500 for Team subscribers.
The pricing lands squarely on OpenAI's turf. At the short tier, Haiku 5.5 matches GPT-6 Luna's $0.10 and $0.50 rates, and Anthropic's own benchmarks put Haiku 5.5 ahead of Luna on every shared test: 72.4% versus 48.9% on OSWorld 2.1, and 39.2% versus 16.4% on Terminal-Bench 4.0. Treat those numbers with the usual caution. No independent benchmarks are available yet, so all of the figures come from Anthropic, and Luna's own higher tier only kicks in above 272K tokens at $0.20 and $0.75, which undercuts Haiku 5.5 on mid-length prompts.
Why this matters: the small-model lane is turning into an open price war, and that is good news for anyone building agents. Cheap, fast inference is what makes multi-step agentic workflows, subagents, classification, and compaction at scale economically sane. Anthropic is explicitly positioning Haiku 5.5 as a subagent to its bigger models, and the launch page's one customer figure points in that direction: Asana reports more than a 30 percent reduction in latency for task completions and up to 2.5x faster inference per agent turn.
The honest read is that list prices are falling faster than effective prices, and the real contest is shifting from raw intelligence to total cost of ownership per task. Tokenizer changes, effort dials, cache economics, and tier thresholds all feed into the actual invoice. Haiku 5.5 is a serious entry in that contest, and developers who run workloads at volume should take a fresh look at the math. Just read the whole pricing table before budgeting.