Anthropic released Claude Haiku 5.5 on 7 October 2026. It calls it the cheapest, fastest and most capable small model it has released. For anyone running an AI agent at volume, the headline is the price.
Price
For prompts up to 100,000 tokens, the listed price is $0.10 per million input tokens and $0.50 per million output tokens. Longer prompts are priced at $0.50 and $2.50. The previous small model, Haiku 4.5, was listed at $1 and $5. Anthropic says Haiku 5.5 costs around 75 percent less to run on average. The API model ID is claude-haiku-5-5.
What Anthropic reports
Anthropic reports a 72.4 percent score on a subset of OSWorld 2.1, a test of using a computer, against 15.7 percent for Haiku 4.5. That is Anthropic's own figure, so run your own tests before relying on it. Still, the size of the jump is why people are looking again at small models for agent work that used to need a larger one.
Good jobs for a small model
- Triage: reading incoming email or enquiries and labelling them.
- Extraction: pulling dates, amounts and names from invoices, forms and messages.
- First-line customer answers: answering from your own FAQ and passing hard cases to a person.
- Summaries: meeting notes, call logs, long documents.
- Routing: deciding which tool or which larger model should handle a task.
What to watch
Small models are more likely to miss a subtle instruction or give a confident wrong answer on an unusual case. Build checks into the agent: a confidence rule, a second look by a bigger model for risky actions, and a human approval step before anything is sent to a customer or money moves.
A simple cost picture
An agent that handles 1,000 enquiries a month, with a few thousand tokens in and out per enquiry, uses a few million tokens. At Haiku 5.5 prices that is a very small model bill. The costs that matter are your own time to set it up, the integrations, and the review of its work. That is why price per token should not be the only deciding factor.
Sonnet 5.5 sits between Haiku and Opus. Anthropic's Haiku page lists it at $2 per million input tokens and $10 per million output tokens for prompts up to 100,000 tokens. Many agents use all three, each for the part of the job it suits.
Want an AI agent built around your own business, with the right model for each job? Talk to Power Digital and we will scope it with you.
