TL;DR

Anthropic's Opus 5.5 launch separates reductions in input, output and cache-read rates from a broader claim about typical task-cost efficiency. Those percentages describe different things and should not be combined into a single guaranteed discount. Estimate costs from your actual input, cache and output mix, and hold the processing mode constant when comparing alternatives.

What changed

Anthropic launched Claude Opus 5.5 with standard rates of $4 per million input tokens, $20 per million output tokens, and $0.20 per million cache-read tokens. Compared with the Opus 5 prices shown in the announcement, input and output rates fall by 20%, while cache reads fall by 60%.

The company's separate claim of 40% lower typical task cost includes efficiency under its default test settings. It is not the same as the token-price reduction. The launch also lists a more expensive fast mode, so comparisons need to hold the processing mode constant.

What it means for your workflow

SaveMyToken editorial take

For a long coding session, separate uncached input, cache reads, cache writes, and output. A single headline discount will not accurately predict the bill.