Claude Haiku 5.5 is the small, fast end of the Claude 5.5 family and a large step up from Haiku 4.5 (Terminal-Bench 4.0 39.2% vs 0%, OSWorld 72.4% vs 15.7%, HLE 45.9% vs 10.2%), while costing a tenth as much. It is the first Haiku model with adaptive thinking and an effort setting, and suits summaries, context compaction, database queries, live support and browser automation, or acting as a fast sub-agent for Sonnet 5.5 and Opus 5.5. Pricing is tiered by prompt size: $0.10/$0.50 per million tokens up to 100K prompt tokens and $0.50/$2.50 above. It accepts text and images with a 1M-token context window and up to 128K output tokens.
| Benchmark | Score | Type | Recorded |
|---|---|---|---|
| SciCode | 55.0 | accuracy | today |
| LCR | 82.7 | accuracy | today |
| TerminalBench | 32.8 | accuracy | today |
| Humanity's Last Exam | 44.4 | accuracy | today |