Don’t figure in costs for AI jobs and then, when you see “cut 40%,” automatically slash your quote by 40% too.
On September 22, Anthropic released Opus 5.5: standard API pricing is $4 per million input tokens and $20 per million output tokens, which is 20% lower than Opus 5. Cached reads are $0.20, which is 60% lower. The so-called “40% lower running cost” refers to the official typical task and results from tests using default settings—it also includes changes in usage.
If you’re doing automation services as an individual, you can run the same job with the new model and separately track uncached input, cached reads, output, and retries, plus manual acceptance and ongoing maintenance. How much less you’ll pay in model bills depends on your specific tasks—you can’t directly treat it as a profit increase.
Source: Anthropic “Introducing Claude Opus 5.5,” September 22, 2026. Not personally tested by the author.
On September 22, Anthropic released Opus 5.5: standard API pricing is $4 per million input tokens and $20 per million output tokens, which is 20% lower than Opus 5. Cached reads are $0.20, which is 60% lower. The so-called “40% lower running cost” refers to the official typical task and results from tests using default settings—it also includes changes in usage.
If you’re doing automation services as an individual, you can run the same job with the new model and separately track uncached input, cached reads, output, and retries, plus manual acceptance and ongoing maintenance. How much less you’ll pay in model bills depends on your specific tasks—you can’t directly treat it as a profit increase.
Source: Anthropic “Introducing Claude Opus 5.5,” September 22, 2026. Not personally tested by the author.