What was announced
Anthropic released Claude Haiku 5.5 on 7 October, positioning it for frequent, narrowly scoped work such as summarising text, categorising requests and retrieving information. It includes an adjustable effort setting. Anthropic also reduced Sonnet 5.5 cache-read prices.
The company reports that Haiku 5.5 costs around 75% less to run than Haiku 4.5 on average. This is a vendor estimate, not a guaranteed saving for every business. Pricing differs for prompts above 100,000 tokens, and changes in token usage also affect comparisons.
What it means for Australian businesses
The useful question is whether a cheaper model can reliably complete one task you already understand. Try a small set of representative, non-sensitive examples, such as sorting enquiries or drafting internal summaries. Have staff check the results against the current process.
Measure accepted outputs, corrections, response time and total spend. Include the time spent reviewing errors: a low API bill can still produce an expensive workflow.
Start with a controlled pilot
Keep permission to send messages or change records separate from the model’s ability to generate text. Use human approval for consequential actions and agree when a task should be escalated to a person or a more capable model.
This release is a reason to test task-specific model selection, not to replace every existing tool or remove oversight.