Anthropic has released Claude Haiku 5.5, the small and fast end of its range, and the headline is the price. For prompts up to 100,000 tokens it costs $0.10 per million input tokens and $0.50 per million output, which Anthropic says is 90% below Haiku 4.5. Beyond that the rates are $0.50 and $2.50, a 50% cut.
The company puts the average saving at around 75% less to run than Haiku 4.5, below the headline discount because the new model uses slightly more tokens per task. It is the first Haiku-class model with an adjustable effort setting, trading cost against capability, and is available on Anthropic’s API and through Amazon Web Services, Google Cloud and Azure.
The gains over Haiku 4.5 on agentic work are enormous, which says as much about the old model’s weakness as the new one’s strength. On Terminal-Bench 4.0, Haiku 5.5 scores 39.2% against 0.0% for Haiku 4.5. On OSWorld 2.1’s offline subset it reports 72.4% against 15.7%, and on Humanity’s Last Exam without tools 45.9% against 10.2%.
Anthropic is direct about the ceiling: Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, and the company says Sonnet and Opus remain the right choice for complex agentic coding. Cache reads on Sonnet 5.5 are also cut to $0.10 per million tokens. All figures are Anthropic’s own.
Source: Anthropic, Introducing Claude Haiku 5.5.