Opus 5.5 and the maths of what to automate
According to Anthropic, Claude Opus 5.5 costs 40% less than Opus 5 on typical workloads and responds faster. A good moment to review the tasks you didn't automate because the numbers didn't work.
Anthropic released Claude Opus 5.5 on 22 September. According to Anthropic, it costs 40% less than Opus 5 on typical workloads and generates output more than 30% faster. It's already available through the API and on AWS, Google Cloud and Azure.
Part of the saving is the price: $4 per million input tokens and $20 per million output tokens, 20% less than Opus 5. Cache reads are 60% cheaper, and according to Anthropic they make up most of the cost of agents and coding. The rest comes from using fewer tokens per task.
An agent's cost per task is a product variable: it sets what you automate and what a person keeps doing. When that cost drops, the line moves.
So make two lists. The tasks you send to a small model purely on price. And the ones you left out because the numbers didn't work. Run them through Opus 5.5 with your own data and check two things: cost per task and the quality of the output. A benchmark doesn't know your product.
You don't need to migrate everything. Start with one task from each list.