What generating a thousand words of AI text costs.
Per-article the cost is trivial; at thousands of articles it becomes a real line item, and prompt size matters. Cut the repeated prompt with prompt caching.
This calculator estimates the dollar cost of generating 1,000 words of AI text for a given model. You enter three values: the prompt tokens per request, the input price per 1M tokens, and the output price per 1M tokens. It converts 1,000 words into roughly 1,333 output tokens, prices that generated text at your output rate, and adds the cost of your prompt tokens billed at the input rate. The result is one per-1,000-words figure, letting you compare models on the same footing and predict what a batch of articles, summaries, or replies will cost before you run them.
The key trade-off is that output tokens usually cost several times more than input tokens, so a long prompt is cheap while the generated text is where spending concentrates. Trimming a bloated prompt helps a little, but picking a cheaper model and keeping responses concise saves far more. Re-run the numbers whenever a provider changes prices or you switch models, since a rate that looks trivial per request compounds quickly across thousands of requests.
About 1,000 words is ~1,333 output tokens. Cost = prompt tokens × input price + output tokens × output price. On a mid model that's a fraction of a cent to a few cents per 1,000 words.
Because you pay for the instructions and any context you send, not just the words generated. A long system prompt on every request adds up across a lot of content.