Open Source Initiative revenue was USD 1,093,077 in 2024, and fell 38.9% in 2025

Quick answer (September 2026). The Open Source Initiative reported revenue of $1,093,077 for calendar 2024, $811,527 for 2023 and $739,664 for 2022 on its Form 990. For 2025 it has published a figure the IRS does not have yet: $667,000, down 38.9%, against expenses of $1,129,000 and a net deficit of $462,000. Contributions were 96.6% of 2024 revenue. Membership dues, paid by the individual members who elect part of the board, were 2.35% of it.

Two of those numbers are already public in some form. The rest are not, and neither is the thing that makes them interesting: OSI publishes its accounts on one basis and files them with the IRS on another, and the gap between the two has been running for at least a decade.

Everything below comes from three places. The FY2023 and FY2024 Form 990s, downloaded as filed XML from the IRS exempt-organisation archive rather than read off any aggregator. OSI's own 2025 annual report, which carries a financial section prepared on 2 March 2026. And the Alfred P. Sloan Foundation's grants database, for one grant that turns out to matter more than its size suggests.

The 2025 number nobody has filed yet

The FY2024 return was signed on 31 October 2025 and transmitted on 3 November 2025. The FY2025 return is not due until late 2026. So on the ordinary timetable, the most recent thing anyone can say about the Open Source Initiative is a calendar-2024 figure.

OSI has gone ahead and published 2025 anyway, in a table whose heading states that it covers consolidated financials in thousands and was prepared on 2 March 2026 by Phyllis Dobbs. It is unaudited. It is also the only place the year exists.

Sponsorships fell 31.3%. Grant income fell 79.9%. Revenue fell 38.9% while spending rose 9.1%, and the resulting deficit of $462,000 is roughly 69% of the whole year's revenue.

OSI does not present that as an accident. Its own text says the board "made an intentional, strategic choice to deploy our accumulated operational reserves", and attributes the revenue drop to "broader macroeconomic contractions in tech sponsorships and the natural operational pause of an Executive Director transition". The reserves did move: net assets fell from $869,000 to $406,000, a 53.3% drawdown in one year, and unrestricted net assets fell 59.6% to $241,000.

Two right answers for 2024

Put the annual report's 2024 column next to the 2024 Form 990 and they disagree on nearly every line.

Neither document is wrong. The difference is a basis difference, and it reconciles to the dollar at the resolution the annual report publishes.

The annual report shows accounts receivable of $68,000 and accounts payable of $28,000 at the end of 2024. The Form 990 shows accounts receivable of zero and accounts payable of zero. Take the receivable out of the report's assets and you get $1,038,000 minus $68,000, which is $970,000, against the return's $970,107. Take the payable out of its liabilities and you get $169,000 minus $28,000, which is $141,000, against the return's $141,058. The net asset gap of $40,000 is exactly the receivable minus the payable.

The return says so itself, in one line of Schedule O explaining the change in net assets: "ADJUSTMENT RELATED TO ACCRUAL TO CASH RECONCILIATION". OSI keeps its books on an accrual basis and files its return on a cash basis.

The receivable the IRS has never seen

That is not a one-year artefact. The annual report also publishes an eleven-year asset table, 2015 to 2025, split into cash, receivables and other assets. Lining its cash column up against total assets on the corresponding Form 990:

In nine of the ten overlapping years the return's total assets match the report's cash line to within $1,000. The tenth, 2020, comes inside $1,700 once you add the $8,000 of other assets the report shows that year.

The practical consequence is 2021. The Form 990 for that year reports total assets of $533,392. OSI's own books put them at $816,000, because $282,000 of it was money owed and not yet received. Anyone reading the return alone saw 65.4% of the organisation's assets.

I read both the FY2023 and FY2024 returns line by line and the receivable is zero at both ends of both years. For the earlier years the match between the two columns is strong evidence of the same treatment rather than a direct reading, and I am stating it as that.

A public support percentage that moved 51.5 points

Schedule A is where a public charity proves it is publicly supported rather than a private foundation in disguise. The test is a five-year rolling average, and the number has to clear 33.33%.

OSI's FY2023 return reports 46.738%. Its FY2024 return reports 98.279%.

The five-year gift totals barely moved: $2,823,496 in the 2019-2023 window, $3,175,717 in the 2020-2024 window. The entire swing sits in one line. Schedule A requires you to strip out, from each substantial contributor, whatever they gave above 2% of total support. In the FY2023 filing that exclusion is $1,495,023, which is 52.95% of every gift OSI received across those five years. In the FY2024 filing the element is not merely zero. It is absent.

Here is why that is worth saying out loud rather than filing away. Two per cent of OSI's FY2024 total support is $64,626. The Sloan Foundation's own grants database records a single grant to OSI of $249,150 in 2024, and Sloan's own 2024 audited financial statements carry the same figure. That one grant, from a private foundation, exceeds the 2% ceiling by $184,524 on the return's own numbers. The return reports nothing excluded.

I am not going to adjudicate that, because the schedule that would settle it is legally withheld. Every field of OSI's Schedule B, the contributor list, reads "RESTRICTED" in the public file. There are two readings. Either the donor base genuinely diversified enough that the test computes differently, or the two filings applied the substantial-contributor calculation differently. The return does not say which, and the document that would tell you is the one the public inspection rules remove.

What can be said cleanly is that OSI passes either way. Apply the FY2023 exclusion in dollars to the FY2024 numbers and the percentage is 52.0%, still comfortably over 33.33%. The status was never in question. It is the visibility that changed.

One more thing in the same table. The 2022 gift figure was restated. The FY2023 filing reports $739,259 for 2022. The FY2024 filing reports $563,897 for the same year, a reduction of $175,362, or 23.7%. The Form 990's own prior-year column was not restated: I checked nine Part I line items across the two returns and every one matches exactly. Only Schedule A's version of 2022 moved. And the restatement works mildly against OSI, not for it. Leave 2022 alone and the FY2024 percentage would have been 98.368% rather than 98.279%.

Who funded the Open Source AI Definition

The Sloan grant is named in OSI's own return, in the last place most people would look. Schedule F, foreign activities, exists on the FY2024 filing and did not exist on FY2023. Its narrative field opens:

Schedule F, Part III, type of assistance: "IN 2024, OPEN SOURCE INITIATIVE RECEIVED A SIGNIFICANT GRANT FROM THE ALFRED P. SLOAN FOUNDATION TO CREATE AND EDUCATE THE PUBLIC ABOUT THE OPEN SOURCE AI DEFINITION"

Sloan's public purpose line for grant g-2024-22486 is "To support the development of an Open Source AI Definition supported by a diverse set of stakeholders". The amount, $249,150, is essentially the entire $249,000 grant line in OSI's own 2024 accounts.

The schedule then does something unusual for a policy organisation. It reports cash grants to 17 foreign individuals: 14 in Europe totalling $24,903, disbursed by PayPal and wire transfer, and 3 in sub-Saharan Africa totalling $1,575, by wire. Not grants to organisations. Individuals, paid directly, averaging $1,558 each.

There is a loose end. Schedule F totals $26,478. Part IX line 3, the foreign grants line, reports $23,332, and Part I agrees with Part IX. The $3,146 difference is not explained anywhere in the return, and because FY2023 has no Schedule F at all there is no prior year to test it against. I am reporting both numbers rather than picking one.

What a million-dollar standards body actually spends

OSI employed one person in 2024, down from two in 2023. Total salaries and benefits were $202,140. Executive director Stefano Maffulli's reportable compensation was $188,678, which is 93.3% of the entire staffing line, up from 81.0% the year before after a 13.2% rise. The remaining $13,463 is payroll taxes: no other salaries, no pension contributions, no other benefits.

The work is bought, not employed. The other-professional-fees line, Part IX line 11g, came to $416,462, which is 2.06 times the payroll, up 61.7% year on year. Schedule O says what that buys: "STANDARDS WORK GROUP, POLICY PROGRAM SUPPORT, TECHNICAL WRITING FEES".

And the Open Source AI Definition year shows up hardest in the travel lines. Travel went from $47,146 to $132,973, a factor of 2.82. Conferences and meetings went from $32,735 to $72,170, a factor of 2.20. Together, $79,881 became $205,143. A global consultation is an airfare line item.

Program spending rose from 57.2% to 64.7% of total expenses, which is the direction you want, and fundraising costs actually fell from $134,209 to $119,210.

On reserves, OSI ended 2024 in good shape and it is worth saying so as plainly as the 2025 drawdown. Net assets of $829,049 against $1,013,797 of annual expenses is 298.5 days of cover, or 200.2 days counting only the unrestricted portion. On the 2025 self-published figures that falls to about 131 days total and 78 days unrestricted. Still solvent, much thinner, and deliberately so.

The only liability on the balance sheet, in both years, is a single unsecured note: $148,470 at the start of 2023, $144,815, then $141,058. No accounts payable, no deferred revenue, no grants payable. The annual report names it: Loan, SBA. It is being repaid at roughly $3,700 a year against interest of about $3,900, and at that rate the balance runs on for roughly another 38 years.

Three small things, since the filings are open

The FY2024 return misspells two sitting directors. It lists "GAIL BLONDELLE" and "SAYEED CHOWDHURY"; OSI's own board page has Gaël Blondelle and Sayeed Choudhury.

It also spells its own principal officer two ways. The FY2023 header reads STEFANO MAFFULLI; Part VII and Schedule J of the same return read STEFANO MAFFULI. The FY2024 return names a different principal officer entirely, Phyllis Dobbs, which is the name that later appears on the 2026 annual report financials.

The return reports executive compensation twice and gets two answers. Part IX line 5 says $188,677. Schedule J and Part VII say $188,678. One dollar.

And OSI's own annual report attributes a 30% increase to its legal and professional fees line. Its own table shows that line going from $473,000 to $655,000, which is 38.5%. The rest of the report's arithmetic is clean: I checked sixteen identities in it and fifteen foot exactly, the sixteenth being program service lines that sum to $631,026 against a stated $631,025.

What the existing answers say

Wikipedia's infobox reports OSI revenue of $811,527 for 2023, cited to ProPublica with an access date of 17 December 2025. That is six weeks after the 2024 return was filed. The single figure manages to be wrong in both directions at once: it is 25.8% below OSI's most recent filed year, and 21.7% above the year OSI has itself published.

ProPublica's Nonprofit Explorer page is the strongest incumbent and deserves credit. It has the 2024 return, renders revenue of $1,093,077, lists every officer, and links the schedules. Its verdict is accurate and current. It is also narrow by design: it publishes the Part I totals and executive compensation, and not the public support percentage, not Schedule A, not Schedule F, not the accrual-to-cash difference, and not 2025. One caveat for anyone scripting against it, since it caught me: the web page carries FY2024, while the API's extracted-filings list still stops at 2023.

OSI's own pages hold five of the first ten Google results for this question, which is unusual and to its credit. But the page that ranks first is an "About" page with no financials on it, and the annual report that has the whole 2025 story sits at position ten.

Method

Both returns were pulled as filed XML by byte-range request against the IRS archive rather than downloaded whole, which is why this took minutes rather than hours. It is the same method behind our piece on Software Freedom Conservancy's revenue. Then I ran the arithmetic.

Sixty structural identities across the two filings reconcile exactly, with no rounding tolerance: revenue composition, the contributions build, the Part IX functional expense footing including every occurrence of the repeatable other-expenses line, of which the 2024 return has four and the 2023 return three, the three expense columns, both balance sheets at both ends, the restricted and unrestricted split, the Part XI roll-forward, and nine prior-year columns tested against the prior filing. Six Schedule A identities reconcile too, including both public support percentages recomputed from their own inputs.

The Part XI roll-forward is worth a note because it did not foot on the first attempt. Both years were out by exactly $800, which is the sort of tidy number that looks like a finding. It was not. It was a line I had failed to read: donated services and use of facilities, $800 in each year. Add it and both years close to the dollar. A failed footing is a statement about your extractor before it is a statement about anyone's accounts.

Joaquin del Rio writes the revenue files. Every figure here is traceable to a filed return, a published report, or a third party's own database, and nothing is estimated.