Claude Haiku 5.5: Anthropic’s fastest small model at a much lower price
The new Haiku starts at $0.10 per million input tokens and gets an effort setting for the first time. Anthropic estimates about 75% savings over Haiku 4.5.
On October 7, Anthropic released Claude Haiku 5.5, the first Haiku-class model since Haiku 4.5. It’s built for high-volume, low-cost work: summaries, database queries, classification, live customer support and browser use.
What’s new
- Price: $0.10 per million input and $0.50 per million output tokens for prompts up to 100k tokens, and $0.50 and $2.50 above that. Anthropic estimates it costs about 75% less on average than Haiku 4.5.
- Effort: for the first time on Haiku, you can pick a level from Low to Max, trading cost against quality.
- Subagents: the model is designed to work alongside Opus 5.5 and Sonnet 5.5 on coding tasks.
- Computer and browser use: in beta through the Python and TypeScript SDKs.
On OSWorld 2.1, Haiku 5.5 scored 72.4% against 15.7% for its predecessor. These numbers come from Anthropic.
How to try it
claude-haiku-5-5 is available on all platforms: the Claude Platform, AWS, Google Cloud and Azure.
Limitations
For complex agentic coding, Anthropic recommends Sonnet 5.5 or Opus 5.5: on Terminal-Bench 4.0, Haiku scored 39.2% against 70.6% for Sonnet 5.5. The updated tokenizer uses slightly more tokens per task than Haiku 4.5, so work out your real savings on your own workload.
Prepared with AI from public sources.