Anthropic said on 7 October that Claude Haiku 5.5, its small, fast model, is out. A token is a small chunk of text, and prices are per million. On the Claude Platform, prompts up to 100,000 tokens cost $0.10 for input and $0.50 for output. Longer prompts cost $0.50 and $2.50. The New Stack reported that Haiku 4.5 costs $1 and $5 at any prompt length. Teams can use the model in Claude.ai and Claude Code, or through Amazon Web Services, Google Cloud and Microsoft Foundry.
Three things set the bill. First, prompt length picks the rate. The New Stack reported that Anthropic says about 90% of Haiku 4.5 requests would fall in the cheaper band. Second, a new tokenizer, which splits text into pieces, uses slightly more tokens per task. The New Stack said Anthropic's average saving of around 75% covers both the request mix and this change. Third, Haiku 5.5 is the first Haiku with effort controls, set to medium by default, which govern how many tokens a task uses. The sources do not say what any one workload will cost or how each effort level changes it.
The per-token price is no longer the whole cost. Buyers can count the tokens their own prompts use, check what share stay under 100,000, and test each effort level. The New Stack noted that Alibaba's Qwen3.7 Flash lists lower prices still.