Anthropic explains Opus 5.5 costs per completed task
On September 22, Anthropic published a task-cost guide for Opus 5.5. It lists API prices of $4 per million fresh input tokens, $20 for output and $0.20 for cache reads.
It describes reductions of 20% for input and output and 60% for cache reads versus Opus 5. Workload-dependent token use still matters; the examples are illustrations, not universal measured savings.
Our analysis
An AI that avoids retries may beat a cheaper one
Low unit prices can be outweighed by repeated failures and rereading. We expect comparison to move from the cost of a reply toward the cost of completed work. Including human review time could reverse apparent rankings.
This raises the value of checks that tell an AI when work is finished. Clear acceptance conditions can reduce rework. People who design good checks may also become the people who lower AI bills.
How could everyday life change?
The following is a possible future based on this news.
Set a finish line and budget before assigning work
A future shop owner might ask an AI to fix a booking form within a set budget, including verification. Tools could be chosen by how many jobs they finish reliably, rather than usage alone.
This is our future scenario. The cost guide offers a useful starting point for thinking about the full price of finishing a job.
