Worked example: $10 blog-draft budget
The default scenario uses a $10 budget, GPT-5.4 mini, 500 input tokens, 750 output words, and 0.75 words per output token. Output tokens per request are 750 / 0.75 = 1,000.
Text request cost is 500 / 1,000,000 x $0.75 plus 1,000 / 1,000,000 x $4.50 = $0.004875 per request.
$10 / $0.004875 buys about 2,051 text requests. At 750 words each, that is about 1,538,462 words or 3,077 pages at 500 words per page.
For GPT-Image-2 planning, 100 text prompt tokens at $5 per 1M plus 8,000 output image tokens at $30 per 1M gives $0.2405 per image. A $10 budget buys about 41.6 images before taxes, minimums, retries, or failed generations.
How the words-per-dollar estimate works
The calculator turns output words into output tokens, prices one text request from input, cached input, and output token rates, then divides the budget by the per-request cost. Pages are output words divided by the selected words-per-page assumption.
How images are estimated
The image estimate uses GPT-Image-2 text input and image output token prices from the OpenAI pricing page. Image token count is an editable assumption because size, quality, and model accounting affect the final token count.
What this estimate excludes
The result excludes taxes, failed generations, retries, moderation calls, storage, vector search, web search, workflow platform tasks, staff review, and post-processing. Add a buffer for real production batches.
What to do next
Run a small sample, export usage logs, replace the token assumptions with actual medians, and set budget alerts before generating at scale.
