Anthropic's New Cheap AI Model: 75% Less to Run, Does Small Jobs

Anthropic just released Claude Haiku 5.5, a new AI model that's much cheaper to run than its previous version. For developers, the cost is as low as 10 cents per million input tokens—that's 75% less than Haiku 4.5. This means businesses can now use AI for routine jobs like sorting emails or summarizing documents without breaking the bank.

The model is also smarter. On a test called OSWorld, it scored 72.4% compared to just 15.7% for the old version. That's a huge jump. It's designed to handle the small, repetitive parts of larger AI workflows—like when a big AI assistant needs to do many little tasks. One customer, Asana, reported 30% faster task completions and up to 2.5 times faster processing per agent turn.

But there's a catch: the lowest price only applies to requests under 100,000 tokens (about 75,000 words). Larger requests cost more, though still 50% less than before. Also, this model isn't meant to replace bigger, smarter models for complex coding or reasoning. Instead, it's for the simple, high-volume work that used to be too expensive to automate.

Anthropic also cut prices for its Sonnet model's cache reads by half and is giving monthly API credits to Max and Team subscribers. Overall, this makes AI more accessible for businesses that want to automate repetitive tasks without spending a fortune.