Output price
What the maker charges per million tokens the model generates, including reasoning tokens. It is usually several times the input price.
Why it matters. Dominates the bill for generation-heavy work like writing code or long answers.
How it works
The output price is the rate for every token the model generates in its reply, quoted per 1 million tokens. It is usually higher than the same model's input price, because generating tokens one at a time takes more compute than reading the prompt.
Reasoning tokens are billed as output tokens, whether or not the API shows you the reasoning. Output rates also vary by service tier and, on some models, by prompt length.
For code generation, long-form writing and reasoning-heavy tasks, output usually dominates the bill. Capping reply length, asking for diffs instead of whole files and lowering effort on easy requests all reduce it.
Worked example
At an illustrative $15 per 1M output tokens, a reply of 2,000 visible tokens plus 6,000 reasoning tokens costs 8,000 / 1,000,000 x $15 = $0.12. The visible part alone would be $0.03.
Common mistakes
- Budgeting only for visible answer tokens and missing reasoning tokens on the bill.
- Asking for whole rewritten files when a diff would do.
- Choosing a model on input price when the workload is generation-heavy.
Across the catalog today
Cheapest output today (per 1M tokens)
See it on a real card: Pricing & cost on Claude Opus 5.5.