Expansion corpus
OpenAlex’s works come in two corpora. The core is the curated catalog most people mean by “OpenAlex” — about 323 million works built and matched from Crossref, MAG, PubMed, DataCite, repositories, and more (295 million of them in the default view; the rest are pieces of other works). The expansion is a rawer layer of about 122 million additional works — mostly datasets and single-repository records — minted once, at the Walden cutover, from records that didn’t match anything in the core. Together they’re the all corpus: about 445 million works.
The expansion was added in the November 2025 Walden update, and for a while carried the name XPAC (“Expansion Pack”). It was a one-time batch and it hasn’t grown since: new works go straight into the core whatever source they come from. You’ll still see “XPAC” in older docs, blog posts, and the is_xpac field name — it means exactly this expansion corpus.
The one thing to know: by default you only see the core, without pieces of other works or pulled works. Every works query excludes the expansion unless you ask for it, so if a count looks surprisingly low (or suddenly doubles), the corpus is usually why.
Core, expansion, all
Think of it as three views over the same pile of works:
| Corpus | What it is | Approx. size |
|---|---|---|
| core | The curated catalog. This is the default. | ~323M works (~295M in the default view) |
| expansion | The added layer — mostly datasets & repository records (the former “XPAC”) | ~122M works |
| all | Core + expansion together | ~445M works |
The corpus selector applies to works only — the other entity types have no expansion, so selecting one there does nothing.
In the REST API
The default is core. Pick a different view with the corpus parameter on any works request:
# Core only (default): ~295M works
curl "https://api.openalex.org/works"
# All works, core + expansion: ~445M works
curl "https://api.openalex.org/works?corpus=all"
# Only expansion works
curl "https://api.openalex.org/works?corpus=expansion"
corpus composes with everything else — ?filter=has_abstract:true&corpus=all counts abstracts across both corpora. It’s a works-only parameter; the other entity types reject it like any unsupported parameter. Each work also carries an is_xpac boolean, so on a corpus=all request you can still tell which results came from the expansion.
Deprecated legacy controls: older docs and code express the same choice with
include_xpac=true(equivalent tocorpus=all) and theis_xpacfilter (filter=is_xpac:true&include_xpac=trueis equivalent tocorpus=expansion). Both still work, but they’re deprecated — usecorpus=in new code. Don’t mix the two vocabularies: a request that combinescorpus=with either legacy control returns an error.
In OQL
In OQL, the corpus goes in parentheses right after works, in OQL’s general form (the underlying selector is corpus, with values core, expansion, and all). With nothing said, a query runs on the core:
get works about kelp
get works (expansion corpus) where published in 2020 and title-abstract-keywords has (kelp)
get works (all corpora) where title-abstract-keywords has (kelp)
The first is the core only; the second, only the expansion; the third, core and expansion together.
Pieces of other works are left out too
The default view also leaves out works that are a piece of another work and are not a traditional scholarly work: a book’s front matter and index, an encyclopedia entry, a figure or data file with its own DOI. The rule is exact and uses two public filters: a work is left out when has_parent_work is true and is_traditional is false. So chapters, proceedings papers and articles stay, because they are traditional works. Where pieces nest (a file in a dataset in a collection), only the top one shows.
Nothing is deleted. Add include_parts=true to any works query and the pieces come back:
# The core corpus with every piece: ~323M works
curl "https://api.openalex.org/works?include_parts=true"
# One work's parts: front matter and chapters alike
curl "https://api.openalex.org/works?filter=parent_works.id:W3109397498"
# Exactly the works the default view leaves out for this reason
curl "https://api.openalex.org/works?filter=has_parent_work:true,is_traditional:false"
A filter on has_parent_work, parent_works.id, parent_works.link_type or lineage shows the pieces too, since it asks about them. Neither include_parts nor these filters bring in expansion works: if a work’s parts are expansion works (as Zenodo records usually are), add corpus=all too. A single work fetched by ID (/works/W…) always answers. Counts on other entities (an author’s works_count) still include the pieces. See Parent and child works.
Pulled works are left out too
The default view also leaves out pulled works: works their publisher, host or authors retracted, withdrew or removed (is_pulled). This rule ignores the corpus: no corpus value brings them back. Nothing is deleted. Add include_pulled=true to any works query and they come back:
# The default view with pulled works
curl "https://api.openalex.org/works?search=ivermectin&include_pulled=true"
A filter or group_by on is_pulled or is_retracted asks about pulled works, so it brings them back too, as include_pulled=true would; include_pulled=false keeps them out even then. A single work fetched by ID always answers, and counts on other entities (an author’s works_count) still include pulled works. See Pulled works.
Why the expansion is opt-in
Two reasons. Data quality: expansion works have thinner, noisier metadata on average — many are bare repository or DataCite records with little more than a title and an identifier. It’s well below the core and improving over time, as we bring the batch up to core standards. Stability: silently doubling everyone’s result counts overnight would have broken a lot of queries and dashboards, so the expansion sits behind a flag you turn on deliberately.
If you’re doing careful bibliometrics or want the well-described literature, stay on the core. If you’re casting the widest possible net — hunting for a specific dataset, or measuring total coverage — reach for all. For where these records come from in the first place, see Sources › Repositories and How it’s built.
Last updated
View as Markdown