I want review to find problems that change whether the work can be accepted. A huge list of suggestions can take more time than the implementation, especially when every comment introduces another preference. Paying less for the reviewer helps, but I still need to know what job that second reading is doing.
I already use Sonnet to implement and Opus to review. Opus 5.5, announced on September 22, lowers the price of that second step.
Anthropic announced input at $4 and output at $20 per million tokens, with cache reads at $0.20. The company estimated typical cost 40% lower than Opus 5, through pricing and token efficiency. I like having more room in the budget for a careful second reading. A lower price only pays off if the reviewer finds what matters.
My reviewer should receive the original request and the changes. The implementer's explanation helps locate decisions, but can carry the hypothesis that needs examination. If both get only the same summary, switching models leaves much of the bias in place. I want review to return to the expected behavior and look for what the diff leaves out.
I also need to distinguish defects from preferences. A finding that the change breaks when the request arrives with an empty list blocks acceptance. A finding that the function would read better split in two is taste, and it can wait. When the reviewer makes every suggestion compulsory, the original assignment never ends and the implementer gets another implementation to do. That is review cost too, and so is the back-and-forth needed to make sense of vague comments.
Model choice should follow the task, and that goes for review too. A small fix and a change with many dependencies deserve different attention. I keep the Sonnet and Opus split while it fits the work coming in, and I will revisit it when conditions change.
I will evaluate Opus 5.5 on bounded tasks by asking one thing of each finding: the condition where the problem appears and why it affects the request. Cost includes the return trip to an acceptable change. Long comments and two agents agreeing do not count as verification; I want evidence that lets me accept the change or correct a specific defect without opening an endless discussion of taste.