Anthropic debuts Claude Haiku 5.5 with massive price cuts

Anthropic has launched Claude Haiku 5.5, introducing its speediest and most economical small-scale model to date. The system targets high-volume, budget-conscious operations such as database queries, text summarization, classification, and live customer support. On average, the new model costs about 75 percent less than its predecessor, Haiku 4.5. For prompts under 100,000 tokens, which represent roughly 90 percent of previous Haiku requests, prices drop by up to 90 percent, though prompts exceeding that limit cost five times more. However, because Haiku 5.5 uses an updated tokenizer that consumes slightly more tokens per task, actual real-world savings may be slightly lower than the raw per-token price cuts suggest. This tokenizer behavior mirrors the Opus 4.x models, which saw token usage rise by roughly 30 percent from tokenizer changes alone.

The model delivers substantial performance gains over Haiku 4.5 across multiple benchmarks. On the GDPval-AA v2.1 knowledge benchmark, Haiku 5.5 scored 1,620, more than doubling the 735 scored by the older model. On Humanity's Last Exam, it reached 45.9 percent without tools and 57.4 percent with tools, up from 10.2 percent and 18.7 percent respectively. It also showed massive improvements in computer use, scoring 72.4 percent on OSWorld-2.1 compared to the previous 15.7 percent. On the Terminal-Bench 4.0 agentic coding benchmark, Haiku 5.5 scored 39.2 percent, whereas Haiku 4.5 scored zero. Anthropic noted that Haiku 5.5 outperformed OpenAI's budget model, GPT-6 Luna, across all tested categories, though it still trails Anthropic's larger Sonnet 5.5 model.

Haiku 5.5 is the first model in its class to offer adjustable reasoning levels, allowing users to balance operational costs against output quality. It is available immediately on Amazon Web Services, Google Cloud, and Microsoft Azure. Alongside this release, Anthropic slashed cache read costs for Sonnet 5.5 by 50 percent, dropping the price from $0.20 to $0.10 per million tokens. The company is also distributing monthly API credits to subscribers, offering $100 for Max-5x, $200 for Max-20x, and up to $500 for Team subscribers to encourage experimentation. Additionally, the company updated its Python and TypeScript SDKs to include beta support for computer and browser use.