Two-thirds of last year's research is free to read. Nearly half of India's is not, and that is the wrong half to be missing

For two weeks this column has been cataloguing what is free: national libraries, open textbooks, 14,702 journals that charge authors nothing. The honest next question is the uncomfortable one. If a researcher used only what is free, what would they not see?

We measured it rather than guessing. On 22 September 2026 we queried OpenAlex — an open index of the scholarly record — for works by publication year and by whether a free full-text copy is locatable.

One paper in three. In India, nearly one in two

Of 14,673,699 works published in 2025, 10,021,346 have a free copy — 68.3 per cent. 4,652,353 are closed: roughly one paper in three.

Now restrict it to research with an Indian institutional affiliation. 576,486 works from 2025, of which 309,041 are open — 53.6 per cent. Which means 46 per cent of Indian-authored work from last year is closed, against 32 per cent globally.

That asymmetry deserves to be better known than it is. Indian research is less open than the world average, so the researcher hurt most by a free-only diet is the one trying to read Indian work — on Indian crops, Indian clinical populations, Indian materials and manufacturing, Indian policy. The literature closest to the problem is the literature least likely to be readable.

The free tier thins out the further back you look

The same query across four years, all measured on 22 September 2026:

| Publication year | Works with a free copy | |---|---| | 1995 | 15.7% | | 2005 | 18.6% | | 2015 | 32.5% | | 2025 | 68.3% |

This is the single most practical finding for a research student. Free-only access is close to sufficient for keeping up, and badly insufficient for looking back. A literature review that reaches twenty years into a field is working in a period where five papers in six were never opened. The foundational paper everyone cites, the method described properly the first time, the negative result from 2003 that would have saved you a year — that is where closed literature lives.

"Open" is not one thing, and the largest slice is not the published paper

Breaking 2025 down by how it is open:

• Green — 4,873,307. An author's manuscript deposited in a repository.
• Closed — 4,652,353.
• Gold — 1,813,669. Published open in a fully open journal.
• Diamond — 1,712,457. Open, with no author fee.
• Hybrid — 1,144,490. Open article inside a subscription journal.
• Bronze — 477,423. Free on the publisher's site with no licence, and revocable.

The categories add exactly to the total. The important one is green, the largest single category, because a green copy is usually the accepted manuscript — the text after peer review, before the publisher's typesetting. Same science, but not the version of record: different pagination, sometimes missing the final corrections, and frequently released only after an embargo of six to twenty-four months.

So part of what looks like openness is openness on a delay, in a version you should not quote page numbers from. And bronze — nearly half a million papers — is free today with nothing obliging it to stay free.

Five things a free-only diet actually costs

Recency, in fields where it matters. Embargoes mean the green copy of a paper published this month may appear a year from now. If you work in a field that turns over in months, free-only means working a year behind.

The version of record, and its corrections. Corrections, expressions of concern and retractions attach to the published version. A researcher reading only manuscripts is reading a record that does not update, and citing a retracted paper is a failure nobody forgives because it is trivially checkable.

Back files. See the table. Older literature is mostly closed, and the deeper a review goes, the more of it is missing.

Standards, handbooks and reference works. Engineering standards, pharmacopoeias, spectral and crystallographic databases, major reference works — these are almost never open, and for an engineering or chemistry group they are not background reading but daily tools.

The discovery databases themselves. Scopus, Web of Science, and the chemistry and engineering indexes are infrastructure rather than content: they are how you find out what exists. Losing them does not remove papers from your reach; it removes your ability to know they were there — the most expensive kind of gap, because it is invisible. We made the same point about AI search tools, which inherit exactly this limitation.

A free-only collection produces a specific, recognisable failure: the reader keeps up with the field and cannot see its foundations. Nothing announces it. The search returns results, the results are current, and the twenty-year-old work that framed the question is simply absent.

What a subscription does not fix

The argument runs both ways, and the sales version of it is wrong.

A subscription does not make the free tier redundant. Indian theses in Shodhganga, agricultural research in Krishikosh, the open Indian society journals, government data and heritage collections are not in any commercial package, and no amount of spending will put them there.

A subscription is also not a substitute for remote access. A package nobody can open from a hostel is money spent on a reading room, which is the mistake we described in setting up a college digital library.

And a subscription bought without usage data is how libraries end up renewing what nobody opens while cancelling what a department depends on.

The order that gets the most out of a fixed budget

Free first, and thoroughly: N-LIST membership, NDLI accounts, the open Indian repositories, browser tools that surface legal free copies, and a deposit policy so your own institution's output joins the open half.

Then measure what is still missing. Which titles do people request? Which does the department cite? Where does interlibrary lending keep going for the same journal?

Then buy against that evidence — consortium routes first, with a perpetual-access clause — and review it annually against COUNTER usage, willing to cancel.

The point of the numbers at the top is not that free access is inadequate. It is that free access is now the majority of current literature and a minority of everything older, which tells you exactly what a subscription budget is for: the closed third, the back files, the standards, and the indexes that tell you what exists.

Why it matters for students and researchers

For a research student, the finding to act on is the back-file one. Do your keeping-up with free tools; expect the historical part of your literature review to need your institution's subscriptions, an interlibrary request, or an email to an author. Budget time for it rather than discovering it in your final year.

For a researcher, the India number is the one to sit with. Forty-six per cent of Indian-affiliated work from 2025 is closed, and every author reading this controls one small piece of that. Depositing your accepted manuscript moves your own paper from the closed column to the green one, at a cost of one afternoon.

For an institution, both numbers are the budget argument in a single line: two-thirds of current literature is free, so buy for the third that is not — and do it against usage data rather than a package brochure.

Frequently asked questions

What is the difference between open access and subscription journals?

An open-access journal makes its articles free to read, usually under a licence permitting reuse. A subscription journal sells access to institutions and individuals. The practical difference for a reader is whether an institutional login is required; the practical difference for an author is who pays and what rights are retained.

Are paid journal subscriptions still worth it for colleges?

For the closed third of current literature, the back files, standards and the discovery databases — yes, if usage data supports the specific titles. For current open literature, no: 68.3 per cent of 2025 works already have a free copy, and paying twice for those is how budgets get wasted.

What are the limitations of free-only resources for a PhD?

Embargo delays on repository copies, manuscripts rather than versions of record, missing corrections and retraction notices, a much thinner historical record — only 15.7 per cent of 1995 literature is open — and no access to standards, reference databases or the major indexes.

Why do institutions subscribe to e-journals at all when so much is free?

Because "so much is free" is true of recent literature and false of older literature, because a third of current work remains closed, and because discovery databases are bought infrastructure rather than content.

Is Indian research more open than average?

No. Of Indian-affiliated works published in 2025, 53.6 per cent have a free copy, against 68.3 per cent globally. Indian research is less open than the world average, which is precisely why Indian repository deposit matters.

How were these numbers produced?

By querying OpenAlex's public API on 22 September 2026, counting works by publication year and by whether a free full-text copy is locatable, and separately for works with an Indian institutional affiliation. The category counts add exactly to the yearly totals, which is a reasonable check that nothing is being double-counted.