Seven AI Drafts of a 1,500-Word Article Cost $0.14 to $1.88. The Writer's Rate for the Same Article Is $375

Why this pays

The question has a small answer and a large one, and a quote that uses only the small one is wrong. The small answer is the token bill for one 1,500-word article through seven drafts. The large answer is the hours you spend beside the model, at a rate a professional body actually publishes. Both belong on the sheet, and the gap between them is the whole business.

The most expensive token bill on the page is a two-hundredth of the cheapest human rate. The client is not paying for tokens; they are paying for the hours you spend checking, and the token line goes on the quote precisely so they can see how small it is.

The play, in three lines

1. Convert one real article into tokens, check which tokenizer your model is on, and run one pass to read the usage line instead of guessing.
2. Multiply by the passes you actually run, seven, not one, at the published rates of Anthropic or OpenAI.
3. Add your editing hours at a rate from the Editorial Freelancers Association's chart, and quote that total, with every prompt and draft saved in your own folder.

The path above walks every step. What follows is the reasoning behind each and the numbers you quote from.

Where the money comes from

• The client, per article, at a human rate. The EFA's 2026 chart: work-for-hire blog posts at 25.0 to 45.0 cents a word or $74.50 to $87.50 an hour; ghostwriting at 15.0 to 30.0 cents a word or $75.00 to $100.00 an hour.
• Your margin, between that rate and the token line. $375 for the article at the lowest blog rate against a token bill under $2.
• Not the provider. They bill per token: Anthropic on prepaid credits with no published minimum, OpenAI from the first call with no plan fee.

The pain you are selling the cure for

The client has heard that AI writes articles for free and wants to know why yours costs $375. The answer on the sheet is the seven passes at $0.14 to $1.88, and then the hours: the fact pass against their sources, the tone pass against their previous pieces, the line edit. A quote that shows both lines answers the question before it is asked, and a quote that hides the token line invites it.

Step one: the job in tokens

• Anthropic publishes two estimates that do not quite agree: its token-counting page says a token approximately represents 3.5 English characters; its pricing page says 1 token is approximately 4 characters or 0.75 words.
• On either, a 1,500-word article is about 2,000 output tokens, and the brief a few hundred input tokens.
• OpenAI's API pricing page publishes a price per million tokens for every model and no character-per-token or word-per-token ratio, so for a gpt model the estimate is Anthropic's and the measurement in step three is what you trust.

Step two: which tokenizer

• Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text; Claude Sonnet 4.6 and earlier use the previous one.
• The same article is 2,000 tokens on one and about 2,600 on the other. Check which side of that line your model sits before you multiply anything.

Step three: one pass, measured

• A Messages API call returns a usage object with input_tokens and output_tokens. Those replace both estimates.
• The rates: Haiku 4.5 $1 per million input and $5 per million output; Sonnet 5 $2 and $10 after the planned rise to $3 and $15 was cancelled; Opus 5 $5 and $25; Fable 5.1 $10 and $50, ten times Haiku for the same tokens. OpenAI: gpt-4o-mini $0.15 and $0.60; gpt-5 $1.25 and $10.00.
• The first chart above is the output column, the expensive half of every bill. One pass of 2,000 output tokens: a cent on Haiku 4.5, two on Sonnet 5, five on Opus 5, ten on Fable 5.1, a tenth of a cent on gpt-4o-mini.

Step four: seven passes, not one

• An outline, a draft, two rewrites, a fact pass, a tone pass and a final: seven passes of roughly the same size. The second chart prices exactly that: $0.14, $0.38, $0.94, $1.88.
• Anthropic offers 50% off with batch processing for work that can wait; OpenAI's batch rows for gpt-5 are $0.625 input and $5.00 output against $1.25 and $10.00 live.
• The drafting guide shows how to write the brief so that seven passes become four.

Step five: the hours, at a published rate

• The EFA's 2026 chart, editing business and marketing copy: $45.00 to $60.00 an hour for proofreading, $50.00 to $60.00 for copyediting, $55.00 to $75.00 for line editing, $60.00 to $75.00 for developmental editing.
• A 1,500-word article at the lowest published blog rate is $375, the last bar on the second chart.
• Your quote is the token line plus your hours at a rate from that chart.

Step six: the folder

• Anthropic auto-deletes API inputs and outputs within 30 days. OpenAI deletes all Customer Content within thirty days of termination and hands nothing back.
• The brief and the seven drafts go into a folder with the client's name on it on the day they are made, or the next revision request means paying for the same words again and arguing about which draft was approved.

What stops the meter

• Anthropic may suspend all access over a suspected Usage Policy breach and disclaims liability for the loss that causes. OpenAI may limit or suspend access for a breach, without prior notice where necessary.
• A publishing schedule that depends on one key has a single point of failure that costs nothing to remove: the second provider's key, in the same folder as the drafts.
• The by-the-word page reads the ownership clauses that let you sell the result.

What to do today

1. Take the last article you wrote. Count its words, convert to tokens, check the tokenizer.
2. Run one pass with its brief and read the usage line. Multiply by seven.
3. Write your hours next to it at an EFA rate. That sheet is the quote, and the token line on it is the smallest number you will write today.