Claude Haiku 5.5: Prices Down Almost 90%, Benchmarks Jump
Anthropic released Claude Haiku 5.5. Prices dropped by nearly 90%, and the computer-use benchmark score jumped from 15.7% to 72.4%.
Anthropic recently announced its latest small model, Claude Haiku 5.5. This model is part of Anthropic's Claude family, and right now it is their cheapest and most compact model. Compared to the previous version, Haiku 4.5, it costs up to 90% less for short prompts, and about 75% less overall. On OSWorld, a benchmark that measures how well an AI can operate a computer on its own, the score improved from 15.7% to 72.4%. On Terminal-Bench, which tests agentic coding in a terminal, the score rose from 0% to 39.2%.
In an agentic setting, the ability to work with a computer efficiently matters a lot. Small, fast models like Haiku usually handle a large volume of simple tasks rather than complex reasoning, so a jump from 15.7% to 72.4% is a big deal. It means a noticeable improvement in task quality for the same money. The model also has a new tokenizer built in, which is what splits input into tokens that the model charges for. This tokenizer produces more tokens per task, so the real-world savings will come out a bit lower than the advertised 90%.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →▌ Related guides

Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles