When a scientist finishes a study, a journal prints it.
For a long time, you had to pay to read it. Your school or your library paid for that. If they did not pay, you could not read the study.
People did not like this. So a new way grew. The writer pays instead, and then anyone can read the paper for free.
That fee has a name. It is called an article processing charge. I wanted to know one simple thing: how big is that bill?
The Bill Is $26.79 Billion
I looked at 510,372,821 research papers. Not a sample. All of them.
Many papers do not say what was paid. But 12,079,570 of them do. Add up just those, and you get $26.79 billion.
That is money from research budgets. Money that did not buy lab time, or pay a student, or run an experiment. It bought the right to be read.
The Strange Part: The Dearest Option Is The Half-Open One
There is more than one way to make a paper free. Two of them cost real money.
The first is a journal where everything is free to read. People call this gold. The middle price there is $2,000 a paper.
The second is a journal that still sells subscriptions. Most of it stays locked. You pay so that your one paper is open. People call this hybrid. The middle price there is $3,250 a paper.
So the half-open way costs more than the fully open way. The journal keeps its subscription money and takes your fee on top.
Across all papers, hybrid fees add up to $8.53 billion.
Medicine Pays The Most
Split the bill by subject and one is far ahead. Medicine papers account for $8.04 billion, at about $2,294 each.
More Science Is Free Now. But Not The Paid-For Kind.
In 2005, 22.5% of papers were free to read in some way. By 2025 it was 88.3%. That sounds like a win, and it is.
But look at which kind grew. Most of the growth is writers putting a copy in a free library of their own, while the journal version stays locked. That went from 15.1% to 77.0%.
The paid-for kind did not grow. Gold journals were 9.88% of papers in 2020 and 4.14% in 2025.
One warning. A paper can become free later, after it is printed. So the last year or two will keep changing after I looked.
A Label That Argues With Itself
Some journals are free to read and free to publish in. Nobody pays. That label is called diamond.
I found 4,625 papers with the diamond label that also list a fee somebody paid. The paper disagrees with its own label.
But be careful here. That is only 0.030% of diamond papers. And remember that only 2.37% of all papers list a fee at all. So this proves the label can be wrong. It does not prove how often.
How I Read Half A Billion Papers On A Laptop
The data is 725.0 GB. That is far too big to download, and I did not download it.
The files are built so a computer can ask for one slice of one file, the way you open a book at one page instead of reading it front to back. I asked each file one question and kept only the answer.
In the end I moved 2.13 GB over the wire to read 725.0 GB. That is 0.2938% of it. It took half an hour.
Four Things I Cannot Tell You
The real bill is bigger. Only 2.37% of papers say what was paid. My total is a floor, not the whole cost.
I cannot split this by country. The field that should hold the country is empty in this data. I checked. It is there, and it is blank. So I say nothing about countries rather than guess.
My middle prices are bands. I counted fees in $250 steps, so every middle price is the bottom of a step.
This is one day's copy. The data is updated often. A later copy will not match these numbers exactly.
Want to see all the charts and data tables?
View the Full Analysis →