Charity data: where to get it

Every figure on our charity pages comes from filings the IRS publishes free to everyone. This page says which file answers which question — the part the IRS’s own download page leaves out — and shares the sector benchmarks we computed from them.

Search Endaoment’s directory

Endaoment is the rail our gifts travel on. Its public directory lists every organisation it can pay, with the description and website the charity gave it. We read the whole directory on the date shown and made it searchable by the words, not only by cause code, so a charity filed under “unclassified” or “Africa” can still be found by what it does. The dated listing is free to download below.

US only Has a contract on Base

The full listing, dated: testio.givabit.io/endaoment/ (CSV and JSON, one file per read). Source: api.endaoment.org, its public read endpoints.

Start here, depending on what you want

If you want…	Download	Notes
A list of every US charity, with cause codes and revenue	Business Master File (eo1–eo4, eo_xx) ~325 MB · CSV · 1.96M rows	The registry. EIN, name, city, state, NTEE cause code, deductibility, status, and a revenue figure. Start here for almost anything — but note the revenue is not always the latest year.
Whether a charity can receive tax-deductible gifts	Publication 78 data 29 MB · zipped pipe-delimited text	Updated monthly. Absence is not proof of a problem: churches need not apply for recognition, and group-exemption subordinates are listed under a parent.
Whether a charity lost its exempt status	Automatic Revocation list 46 MB · zipped pipe-delimited text	Read BOTH date columns. An organisation revoked and later reinstated appears here with a reinstatement date, and treating that as revoked would be wrong.
One charity's finances, without downloading 150 GB	Index CSV for a year, e.g. index_2025.csv ~65 MB · CSV	The lookup nobody mentions. Maps an EIN to a filing's OBJECT_ID, so you can pull a single return out of a single archive instead of unpacking everything.
The filings themselves	Form 990 series XML, 2019–present ~30 GB zipped · 150 GB unpacked	About 69 ZIP archives holding 4.8 million individual XML files, averaging 33 KB each. There is no single giant XML — this is the only form it is published in.
The smallest charities	Form 990-N (e-Postcard) 89 MB · zipped text	Organisations under the filing threshold. Almost no financial detail — a name, an EIN, and confirmation they still exist.

Things worth knowing before you start.

2019 is the earliest year. The IRS no longer publishes 2011–2018, and the old AWS mirror at s3://irs-form-990 still answers but returns zero objects — it was emptied after updates stopped at the end of 2021.
Only the full Form 990 carries the detail. 990-EZ and 990-PF do not, and they are a large share of all filings, so small charities are always sparser in any analysis than large ones.
Three archives use DEFLATE64, which Python’s zipfile cannot read. Use the unzip binary for those, or you will silently skip them and bias your results toward recent years.
A charity’s newest filing may be a 990-T (unrelated business income), which holds none of these figures. Take the newest return of a type that actually carries financials, not simply the newest return.

Browse by cause code

We also publish the underlying filings split up: one CSV per IRS cause code per year for 2025 and 2026, covering 2,102,002 Form 990 filings. Each row is one charity, with revenue, contributions, expenses, programme spend and highest reported salary.

Browse the cause-code extracts →

Sector benchmarks we computed

Free to use for anything, no attribution required. Built from 728,104 Form 990 filings across tax years 2022–2024, grouped by NTEE cause code. We publish these because the filings are public but the aggregation is work — and one charity’s numbers mean little without knowing what normal looks like in its sector.

ntee-benchmarks-3yr.csv

CSV · 51 KB · 704 cause codes

Quartiles and medians per cause code, averaged over three complete tax years (2022–2024, 728,104 filings). Three years rather than one because a single year catches a capital campaign or a bad market and reports it as the character of the organisation.

ntee-distributions-3yr.json

JSON · 13 MB · full distributions

Every underlying value, sorted, per cause code. Use this if you want a percentile at an arbitrary point rather than the fixed quartiles.

Two cautions about using these. Ratios are not comparable across cause codes — a school charges tuition and shows a 17% donation share, a food bank shows 95%, and neither fact says anything about how well either is run. And the programme expense ratio is a contested measure: it rewards organisations that under-invest in staff and systems, which is why the major charity raters publicly stepped back from it. We publish it because it is on the filing and people look for it, not because we think it decides anything.

We link to the IRS rather than mirroring it, so nobody downloads a stale copy from us. IRS data is a work of the US federal government and is in the public domain. The two files we host are aggregates we derived; the underlying filings belong to the organisations that filed them.