On October 8, Anthropic brought to the forefront the lightest and most powerful member of its family—Claude Haiku5.5 was officially released. The official description of it is clear: this is the company’s fastest and most powerful small model to date, designed specifically for handling high-frequency tasks that are extremely cost-sensitive, such as article summaries, content classification, database queries, customer service responses, browser automation operations, and sub-tasks generated by large AI agents when performing heavy tasks.
This small model’s cutting technique is very efficient. The average operating cost of Haiku5.5 is about 75% lower than that of the previous generation Haiku4.5. For requests containing less than 100,000 tokens, the input and output costs are only 0.1 dollars per million tokens and 0.5 dollars per million tokens respectively—which means the amount of money needed to run a large number of lightweight tasks in the past is now almost negligible. More importantly, the Haiku series is the first to introduce adjustable reasoning intensity: users no longer need to allocate full computing power for every sentence, but can adjust the depth of their thinking according to the importance of the task, allowing cheaper small models to think a few extra steps when needed.
The price mechanism of Sonnet5.5 next door was also changed. Anthropic halved the cache read price for Sonnet5.5, bringing it to $0.1 per million Tokens. According to them, this reduces the operating cost of most of their AI tasks by about 20%. Cache reading is the most frequently called upon process when an Agent system repeatedly reviews long contexts. By making this change, it removes the constraints on those autonomous AI agents that operate 24/7 in the background.
The landing passage is also very wide. Haiku5.5 is now available on four clouds: Claude Platform, AWS, Google Cloud, and Microsoft Azure. Enterprises can integrate it into their existing workflows with little need for additional infrastructure. When Anthropic has reduced the cost of the strongest small models to a minimum, and given them an adjustable brain, the role of small models in the AI era has shifted from a marginal role to the forefront of cost management.