Summaries, database queries and other small AI tasks now have a cheaper option in Anthropic's model lineup. Claude Haiku 5.5 launched on October 7, 2026. It also lets developers adjust how much effort the model spends on an answer.
The intended use is repeated, narrowly defined work. Anthropic suggests using Haiku for classification, summaries and smaller jobs delegated by a larger coding model. The company still recommends Sonnet 5.5 and Opus 5.5 for more complex coding tasks.
For prompts up to 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. Tokens are the units a model uses to process text. Haiku 4.5's listed rates are $1 and $5 respectively, making the new token prices 90 percent lower in that tier.
Longer prompts change the calculation. Above 100,000 tokens, the new rates rise to $0.50 for input and $2.50 for output, per million tokens. Those prices are half the previous model's rates.
Anthropic estimates an average running-cost reduction of about 75 percent. That figure depends on its previous customers' request lengths and changes in token use, so it is not a guaranteed saving for every application.
The adjustable effort setting gives developers another trade-off between cost and capability. Anthropic's published tests show larger models still ahead on demanding coding work; a lower price does not erase that gap.
According to the announcement, Haiku 5.5 is available through the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure. The prompt-length threshold is part of the bill.
