Claude Haiku 5.5 Launches at a Dime per Million Tokens, 90 Percent Cheaper Than Its Predecessor
Anthropic released Claude Haiku 5.5 on October 7, priced at $0.10 per million input tokens and $0.50 per million output tokens, which the company says is 90% cheaper than Haiku 4.5 for requests under 100K tokens.
Three key facts
Pricing has two tiers, and 100K tokens is the line. Under 100K tokens: $0.10 input, $0.50 output and $0.01 for cache reads, all per million tokens. Above 100K, input rises to $0.50 and output to $2.50, a fivefold jump that Anthropic frames as still 50% cheaper than Haiku 4.5 for large requests. The model, ID claude-haiku-5-5, is live on AWS, Google Cloud, Microsoft Azure and the Claude Platform.
The scores leap a grade, and the predecessor nearly handed in blank papers. In Anthropic's own table, OSWorld 2.1 computer use climbs from Haiku 4.5's 15.7% to 72.4%; Humanity's Last Exam without tools goes from 10.2% to 45.9%; Terminal-Bench 4.0 goes from 0.0% to 39.2%. The in-house mid-size Sonnet 5.5 still scores 70.6% on Terminal-Bench, so the gap is real.
Haiku gets an effort dial for the first time. Developers can pick from five reasoning levels: Low, Med, High, Xhigh and Max. Among customers, Asana reports more than 30% lower latency and 2.5x faster inference per agent turn, while Box reports 11 points higher accuracy at about half the latency. Sonnet 5.5 cache reads were also cut by 50%.
WangDou's Take
A dime for a million tokens covers most of the Harry Potter series, with enough change left for a stick of gum. But the fine print is clear: cross 100K tokens and the price instantly multiplies by five. It's the hot-pot joint advertising a one-dollar first plate of beef. The real eaters are all on plate two. Even better is the report card: to make the new model look like it skipped a grade, Anthropic cheerfully printed the old model's 0.0% on Terminal-Bench. That's a dealer selling you this year's car while mentioning that last year's model they sold you never actually started.