OpenAlex: What It Is & How It Works

OpenAlex is a free, fully open catalog of the global research system — works, authors, institutions, and the other entities that make up scholarly output — built and maintained by the nonprofit OurResearch. Unlike Scopus or Web of Science, there is no subscription: the underlying dataset is licensed CC0 and can be queried through a REST API or downloaded in bulk. For research administrators evaluating tools for a CRIS, an institutional analytics dashboard, or a research-impact report, OpenAlex is increasingly the infrastructure layer other systems build on rather than just another search box for researchers.

This guide covers what OpenAlex actually is, the real history of its succession from Microsoft Academic Graph, its core data entities, how its API and bulk-download model work (including a real, recent pricing change), and why its open licensing specifically matters for an institution building its own research-information systems.

What OpenAlex Is

OpenAlex is a bibliographic and scholarly-metadata index — named after the ancient Library of Alexandria — that catalogs scholarly works and links them to the people, organizations, journals, publishers, and funders behind them. It is built and stewarded by OurResearch, a Vancouver-based 501(c)(3) nonprofit founded in 2011 that also builds Unpaywall, the open-access-detection tool many library discovery layers and browser extensions already rely on. OpenAlex’s own funding has come in significant part from the Arcadia Fund, a charitable foundation that has backed several pieces of open scholarly infrastructure.

Two things distinguish OpenAlex from most bibliographic databases a research office has historically paid for:

The data is CC0-licensed. There is no copyright claim on the metadata itself, so it can be downloaded, redistributed, remixed, and built into another product without a license negotiation.
OurResearch is a signatory to the Principles of Open Scholarly Infrastructure (POSI) — a governance commitment covering open data, open source code, and a sustainability model that doesn’t depend on the data itself ever becoming restricted. OurResearch was among the earliest organizations to make this commitment, in 2021, and reaffirmed it after POSI’s version 2.0 update.

From Microsoft Academic Graph to OpenAlex: A Real Succession, Not a Rebrand

OpenAlex exists in its current form because of a specific, verifiable event: in May 2021, Microsoft announced it would fully retire Microsoft Academic Graph (MAG) and the associated Microsoft Academic search service by the end of that year. MAG had been a widely used free bibliographic dataset since 2016, and its planned shutdown left a real infrastructure gap for anyone who had built tooling on top of it. OurResearch had already been developing a MAG-adjacent index; when Microsoft’s shutdown was confirmed, OurResearch incorporated the final MAG data dump (excluding patent data) as a starting corpus and launched OpenAlex on January 1, 2022 — timed to go live essentially as MAG went dark, positioned explicitly as a free, open, ongoing replacement rather than a one-time archive of MAG’s final snapshot.

This history matters practically, not just trivially: any institutional tooling, citation-matching script, or analytics pipeline still referencing Microsoft Academic identifiers is referencing a dataset that has not been updated since 2021. OpenAlex is the actively maintained continuation of that lineage, and its own entity IDs (not MAG IDs) are what current integrations should be built against.

OpenAlex’s Core Entities

OpenAlex organizes everything around a small set of linked entity types, each with its own stable ID and API endpoint. Understanding this entity model is the prerequisite for using OpenAlex as backend data for a CRIS or an analytics build, since it determines what a query can actually join against:

Works — the core scholarly output records: journal articles, preprints, books, datasets, and similar. Each Work carries metadata (title, authorship, publication venue, open-access status) plus resolved links out to identifiers such as a DOI where one exists.
Authors — person-level records, disambiguated algorithmically across name variants and, where available, linked to an ORCID iD.
Sources — the journals, conference proceedings, repositories, and preprint servers that host Works.
Institutions — universities, hospitals, companies, and other organizations, linked where possible to a ROR ID for unambiguous institutional identification. This link is assigned algorithmically from the affiliation text printed on the paper, and it is the single noisiest part of the OpenAlex graph — see the institution-disambiguation section below before using Institutions for any institutional reporting.
Topics — OpenAlex’s current subject-classification system, developed with Leiden University’s Centre for Science and Technology Studies (CWTS). Roughly 4,500 topics are organized into a four-level hierarchy (domain, field, subfield, topic), assigned to Works through a combination of citation-network clustering and machine classification. Topics superseded OpenAlex’s earlier Concepts field, which OpenAlex now documents as deprecated: Concepts values are still returned on older Work records for backward compatibility, but are no longer actively maintained, and new integrations should use Topics instead.
Publishers and Funders — organization-level records for the entities that publish Sources or fund the research behind Works, useful for institutional or funder-level reporting rollups.

Every entity links to the others — a Work points to its Authors, its Source, its Authors’ Institutions, its Topics, and any Funders acknowledged in it — which is what makes OpenAlex usable as a graph rather than a flat list of records.

The API and the Bulk-Data Model

OpenAlex is queryable two ways, and the choice between them matters for how an institution actually builds on it:

The REST API (api.openalex.org) supports filtering, full-text search across abstracts, and pagination across all seven entity types — the right tool for on-demand lookups, such as resolving a single Work by DOI or pulling a researcher’s current output list. A free account and API key (obtained at openalex.org) is the recommended way to query it — CASRAI’s OpenAlex API reference covers the entity endpoints, authentication, filter/search/select syntax and cursor pagination in detail. In February 2026, OpenAlex introduced usage-based pricing for API calls: different operations (a bulk search versus a single lookup by ID or DOI) cost different, published amounts, with every API key receiving a modest daily free allowance and single-record lookups by ID or DOI kept free regardless of volume. OurResearch has been explicit that this is a “free data, paid service” model, not a retreat from open access — the API is a hosted convenience layered on top of data that remains unrestricted.
The full snapshot is a complete, periodically refreshed dump of the entire dataset (hundreds of millions of Works and their linked entities), distributed as newline-delimited JSON and Parquet files in a public, anonymously-accessible Amazon S3 bucket — no AWS account, sign-up, or requester-pays charge required to sync it. This is the path an institution building its own local research-analytics database or CRIS backend actually wants: rather than paying per API call at scale, pull the whole graph once, refresh it on a schedule, and query it locally under the same CC0 terms.

The distinction is the same one that shows up in OurResearch’s own framing: the API now has a metered cost structure for heavy, on-demand use, but the underlying data — the actual CC0-licensed content — remains free to download and self-host in full, with no paywall on the dataset itself.

Why Fully Open Licensing Matters for Research Institutions

This is the practical reason OpenAlex shows up repeatedly in CRIS and research-analytics conversations rather than staying a researcher-facing search tool:

Related reading: Unpaywall. Unpaywall is OurResearch’s free database, API, and browser extension for detecting legal open-access copies of scholarly articles by DOI.

No subscription gate on the underlying data. Scopus and Web of Science require an institutional license just to query their APIs at all, and neither publishes a free, complete, redistributable bulk export of their full metadata under an open license — see CASRAI’s Scopus vs Web of Science vs OpenAlex comparison for how the three actually differ on coverage, cost, and access model. An institution can build a production analytics pipeline on OpenAlex without a procurement cycle.
CC0 licensing removes redistribution risk. A CRIS vendor, an institutional repository, or a university’s own reporting dashboard can ingest, transform, and re-publish OpenAlex-derived data (institutional publication counts, a faculty member’s output list, a funder-acknowledgment rollup) without navigating a data-use agreement that restricts what can be done with the output.
The entity model is built for interoperability, not lock-in. Because Works, Authors, and Institutions resolve out to DOIs, ORCID iDs, and ROR IDs — the same identifier infrastructure euroCRIS-aligned CRIS platforms are already built around — OpenAlex data joins cleanly against records an institution already holds from Crossref or DataCite, rather than requiring a separate crosswalk. See CASRAI’s identifier crosswalk guide for how these systems fit together in practice.
Governance transparency. POSI’s stakeholder-governance principles give institutions a real answer to “what happens if this vendor changes terms or shuts down,” which is a legitimate procurement question for any infrastructure a CRIS depends on — the answer for OpenAlex is closer to “the data ships out the door with you” than it is for a proprietary index.

OpenAlex Compared to Related Tools

OpenAlex gets compared against several different categories of tool, and conflating them is a common source of confusion:

Scopus and Web of Science are subscription, editorially curated citation indexes with narrower title lists but (for Web of Science specifically) a longer, more selective curation tradition. See the full Scopus vs Web of Science vs OpenAlex comparison.
Semantic Scholar is a free, AI-powered discovery and search tool built by the Allen Institute for AI — strong on AI-generated summaries and citation-influence classification, and, like OpenAlex, offers an open API. The two are more complementary than competing: some institutions pull structured metadata from OpenAlex and use Semantic Scholar’s NLP-derived features (like TLDR summaries) as a separate enrichment layer.
Research Rabbit is a citation-mapping discovery layer built on top of open bibliographic graph data rather than an independently curated index in its own right — a different category of tool than OpenAlex (a source-of-record graph) even though both sit in the open-infrastructure ecosystem.
OpenAIRE Graph is the one most often confused with OpenAlex, since both are free, open, CC0-adjacent scholarly graphs. They are genuinely different infrastructure: OpenAIRE Graph is maintained by the European OpenAIRE initiative, is oriented around European Open Science Cloud (EOSC) compliance and harvesting from CRIS systems under OpenAIRE Guidelines, and is distributed via Zenodo dumps. OpenAlex is a US-nonprofit-maintained, globally scoped index with its own independent entity model and ID system. An institution may reasonably use both for different purposes rather than treating one as a substitute for the other.

Coverage and Data Quality Against Scopus and Web of Science

The most common single question about OpenAlex is whether it is “good enough” to replace a Scopus or Web of Science subscription. The honest answer is that the three databases are not competing on the same axis: OpenAlex almost certainly indexes more of your institution’s output than either subscription index does, and it is simultaneously less precise about which output is yours. Coverage and accuracy are separate questions, and the right decision depends on which of the two your use case is more sensitive to.

Why the size difference exists

OpenAlex is substantially larger than Scopus or Web of Science, and the gap is a direct consequence of curation policy rather than of better engineering. Scopus and Web of Science both operate a selective title list: an editorial board evaluates journals against published criteria and admits or removes them, which is why Scopus delisting and Web of Science delisting are formal processes with an appeal path. OpenAlex ingests inclusively from Crossref, PubMed, publisher feeds, repositories and preprint servers, and does not curate a title list at all. Everything that follows — the larger corpus, the wider open-access coverage, the weaker signal about venue quality — falls out of that one difference.

On corpus size, do not trust a number you read in a guide, including this one. OpenAlex reports a corpus in the hundreds of millions of Works and it grows continuously, with a refreshed snapshot published on a rolling schedule. Any specific figure is out of date by the time it is published. If a record count is load-bearing for your analysis — a denominator, a coverage percentage, a report footnote — take it from openalex.org or from the snapshot manifest on the day you run the analysis, and state the date alongside it.

What the extra coverage costs you

The comparative scientometrics literature has converged on a consistent pattern: OpenAlex achieves high recall against a known publication list, while its precision at the level of an individual institution is lower than Scopus’s. In practice that means an OpenAlex query for your institution will find nearly everything you published and will also hand you a meaningful quantity of work that belongs to someone else. See, for example, Reference Coverage Analysis of OpenAlex compared to Web of Science and Scopus (arXiv:2401.16359) and the ongoing comparative work published in Scientometrics. The size of that precision gap varies substantially by institution and by country, so treat published figures as an indication that you must measure it locally, not as a value you can adopt.

Two further quality differences matter for assessment work specifically:

Reference and citation completeness. OpenAlex’s citation counts are built from the reference lists it can actually see. Where a publisher does not deposit open references with Crossref, those citations are invisible to OpenAlex but may still be counted by a subscription index that extracts them itself. This is exactly the gap that OpenCitations and I4OC exist to close, and it is why OpenAlex and Scopus can report different citation counts for the same paper without either being wrong.
Document typing and venue signal. Because there is no curated title list, OpenAlex carries no equivalent of the editorial judgment implied by inclusion in Web of Science‘s indexes. If your assessment framework depends on a venue’s indexed status — and many national assessment exercises do — OpenAlex cannot supply that signal and you will still need the subscription source. CASRAI’s guide on whether Web of Science citation data is reliable for research assessment covers the mirror-image problem on the subscription side.

A decision rule

Use OpenAlex where you need breadth, redistributable data and reproducibility: discovery, open-access monitoring, funder-acknowledgement analysis, building a CRIS backend, or any analysis a third party must be able to re-run without buying a licence. Use Scopus or Web of Science where you need a curated denominator or a defensible venue list: a formal assessment submission, a ranking return, or a benchmark that a peer institution will contest. Many research offices run both and reconcile the difference — and the reconciliation itself is often the most informative output. The full three-way breakdown is in CASRAI’s Scopus vs Web of Science vs OpenAlex comparison, and metadata search engines compared puts both alongside Dimensions, CORE, BASE and Semantic Scholar.

The Institution-Disambiguation Weakness You Have To Correct For

This is the failure mode every serious institutional user of OpenAlex eventually hits, and it is worth understanding before you build a report on top of it rather than after a dean asks why the publication count moved. OpenAlex does not receive a clean institutional identifier from publishers. It receives the affiliation string as printed on the paper — free text, in any language, with any abbreviation, department name, hospital name or postal address the author happened to type — and has to infer a ROR ID from it. That inference is probabilistic, and it is wrong often enough to matter.

How the assignment actually works

Per OpenAlex’s own documentation, raw affiliation strings are resolved to institutions in three stages:

A deep-learning model trained by OpenAlex reads the string and assigns one or more institutions to it.
A monthly string-matching pass — a rules layer that adds or removes affiliations based on the raw string, correcting common model errors. It runs once a month, which means a correction you trigger is not visible immediately.
The ROR affiliation matcher, ROR’s own matching service, integrated into the OpenAlex codebase.

OpenAlex states that the parser reaches roughly 0.92 recall and 0.93 precision on the AffilGood benchmark. Read that carefully: it is a corpus-wide benchmark figure across affiliation strings in general, not a guarantee for your institution. Errors are not evenly distributed, and the concentrations are predictable.

The April 2023 training cutoff — the detail that explains most surprises

The single most useful fact in OpenAlex’s documentation on this subject, and the one most often missed: the model has not been retrained since April 2023. Steps 2 and 3 above exist specifically to fill the gaps it leaves. The direct consequence, in OpenAlex’s own words, is that institutions added to OpenAlex or to ROR after that date “won’t be predicted by the model alone” — they depend entirely on the monthly rules pass and the ROR matcher to be caught at all.

If your organisation registered its ROR ID recently, merged, renamed, spun out an institute, or is a newer institution in a country whose ROR coverage has grown since 2023, this is very likely the explanation for an implausibly low publication count. It is not a general statement that OpenAlex is inaccurate; it is a specific, dated, checkable cause with a specific remedy.

Where errors concentrate

Hospitals, health systems and medical schools. A single clinical paper may print a university, a teaching hospital, a research institute and a department. Which of those OpenAlex resolves determines whose output it counts, and reasonable people disagree about the right answer.
Multi-campus, federated and collegiate universities, where the parent/child relationship is genuinely ambiguous and the affiliation string frequently names only the campus.
Non-Latin scripts and transliteration, where the same institution appears under several romanisations.
Generic institute names that collide across countries — any “National Institute of…” or “Institute of Physics” pattern.
Recently registered ROR records, for the training-cutoff reason above.

The ROR pass-through problem, and why your counts move on their own

OpenAlex follows ROR’s parent/child relationships directly rather than maintaining its own editorial hierarchy the way Scopus does. That is the right choice for transparency and reproducibility — the hierarchy is public, versioned and correctable by the institutions themselves — but it has a consequence research offices should plan for: when ROR changes a relationship, your OpenAlex publication count can change even though no publication did. A newly asserted parent relationship can roll a research institute’s entire back catalogue into a university’s total overnight.

The practical implication is a reporting discipline, not a complaint: for any longitudinal series or any figure that will be compared year on year, pin a snapshot and record its date, rather than re-querying the live API each time and silently reporting a different denominator. CASRAI’s guides on registering and maintaining a ROR ID and on how ROR, ISNI, Ringgold and GRID cross-reference each other cover the upstream record you actually control.

The correction workflow

None of the above makes OpenAlex unusable for institutional analysis. It makes an unverified OpenAlex institution query unusable. The corrective steps, roughly in order of effort:

Always pull raw_affiliation_strings alongside the resolved institutions. OpenAlex preserves the original affiliation text verbatim on every authorship, which is precisely what makes independent correction possible — you can see what the matcher was working from and judge the assignment yourself. Any pipeline that discards the raw string has thrown away its only audit trail.
Build your own affiliation-variant list from those raw strings: the abbreviations, former names, department forms, hospital names and misspellings that genuinely refer to you. Use it as an allowlist to recover missed works and a denylist to drop false positives. This is a one-off exercise that pays for itself immediately.
Reconcile against a source you control — your CRIS, your ORCID-asserted affiliations, your HR roster — rather than against OpenAlex alone. See author name disambiguation for the person-level counterpart of this problem, and the identifier crosswalk guide for joining the two.
Decide your lineage policy explicitly and write it down. Each Institution record carries a lineage of parent organisations. Whether your affiliated hospital’s or spun-out institute’s output counts as yours is an institutional policy decision, not a data question — but it must be a stated, consistent one, or your numbers will not be comparable to last year’s.
Use the Affiliation Editor if you are an OpenAlex member. Announced 19 February 2026, it lets authorised institutional curators link raw affiliation strings to their organisation’s ROR record, search affiliation variants, filter matched against unmatched entries, and add or remove linkages. Two scope limits are worth knowing before you plan around it: it does not let you claim works whose affiliation text does not mention your organisation at all, and it does not model internal hierarchy such as departments or labs. The earlier works-magnet submission pathway has been closed, with prior submissions processed.
State your method in the output. Any report built on OpenAlex institution data should record the snapshot date, the ROR IDs included, the lineage policy, and whether affiliation strings were manually curated. This is what makes the number defensible when it is challenged.

The stakes are not purely bibliometric. Affiliation errors propagate into research-security and collaboration screening, where a spurious institutional link can put an organisation in front of a compliance question it has no actual connection to — a problem OpenAlex has itself acknowledged and acted on publicly, covered in CASRAI’s report on OpenAlex correcting bad affiliation matches feeding research-security watch lists. For what institutional OpenAlex data is genuinely well suited to once corrected, see what OpenAlex is actually used for, and bibliometric analysis methodology for the surrounding workflow.

Frequently Asked Questions

Is OpenAlex actually free?

The underlying data is free without qualification: it is CC0-licensed and downloadable in full as a bulk snapshot with no account, payment, or usage restriction. The hosted REST API is free for typical use (a daily free allowance per API key, and unlimited free single-record lookups by ID or DOI), with usage-based charges applying only to heavy, high-volume search and filter traffic under the pricing model introduced in February 2026.

What replaced Microsoft Academic Graph?

OpenAlex did, directly. Microsoft announced MAG’s full retirement in May 2021, and OpenAlex launched January 1, 2022, incorporating the final MAG data dump as part of its starting corpus and continuing as an actively maintained index going forward. Any tooling still built against Microsoft Academic identifiers is referencing a dataset frozen since 2021.

What license is OpenAlex’s data released under?

CC0 — a public-domain dedication with no attribution requirement, though citing OpenAlex as a source is good practice and is what most published research using it does anyway.

What are OpenAlex Topics, and how are they different from Concepts?

Topics are OpenAlex’s current subject-classification system: roughly 4,500 topics in a four-level hierarchy (domain, field, subfield, topic), built with Leiden University’s CWTS using citation-network clustering plus machine classification. Topics replaced the earlier Concepts field, which OpenAlex has deprecated — Concepts values remain on older records for backward compatibility but are not actively maintained, so new integrations should classify against Topics.

Does OpenAlex have rate limits, and do I need an API key?

An API key (free to obtain from an OpenAlex account) is the recommended way to query the API and is required to get the full daily free usage allowance under the current pricing model. Simple lookups of a Work by its DOI or OpenAlex ID are free regardless of volume; search and filter operations draw against the daily allowance or, beyond that, a prepaid usage balance.

How is OpenAlex different from OpenAIRE Graph?

Both are free, open scholarly graphs, but they’re built and governed independently: OpenAIRE Graph is a European-initiative project oriented around EOSC compliance and CRIS-system harvesting, distributed via Zenodo; OpenAlex is a globally scoped index maintained by the US nonprofit OurResearch, queryable through its own API and AWS-hosted snapshot. They are not interchangeable, and some institutions draw on both.

How does OpenAlex compare to Scopus and Web of Science on coverage?

OpenAlex indexes considerably more works than either, because it ingests inclusively from Crossref, PubMed, repositories and publisher feeds instead of curating a selective title list the way both subscription indexes do. The trade-off is precision rather than breadth: OpenAlex typically achieves high recall against a known institutional publication list while returning more works that belong to other institutions, and it carries no equivalent of the venue-quality signal implied by inclusion in a curated index. See the three-way comparison for the detail.

Why does OpenAlex show the wrong institution on my papers?

Because institution assignment is inferred from the free-text affiliation string printed on the paper, not supplied as an identifier. OpenAlex resolves that string in three stages: a deep-learning model, a monthly string-matching rules pass, and ROR’s own affiliation matcher. The model has not been retrained since April 2023, so institutions added to OpenAlex or ROR after that date are not predicted by the model alone and depend on the other two stages. Recently registered, renamed, merged or spun-out organisations are therefore the most commonly affected, along with hospitals and health systems, multi-campus universities, transliterated names and generic institute names.

Can I correct OpenAlex’s affiliation data for my institution?

Yes. OpenAlex member institutions can use the Affiliation Editor, announced 19 February 2026, to link raw affiliation strings to their ROR record and to add or remove linkages. It does not let you claim works whose affiliation text never mentions your organisation, and it does not model departments or labs. Independently of membership, you can correct your own analysis at any time by pulling raw_affiliation_strings alongside the resolved institutions and applying your own allowlist and denylist — OpenAlex preserves the original text verbatim precisely so this is possible.

Why did our OpenAlex publication count change when we did not publish anything?

Most often because ROR changed a parent/child relationship. OpenAlex follows ROR’s hierarchy directly rather than maintaining its own editorial one, so a newly asserted parent relationship can move an institute’s entire back catalogue into a university’s total with no change to the underlying publications. Pin and date a snapshot for any figure that will be compared year on year, rather than re-querying the live API each time.

How many works does OpenAlex contain?

Hundreds of millions, and the figure rises continuously as new snapshots are published — which is why no specific count should be quoted from a secondary source, including this guide. Take the current number from openalex.org or the snapshot manifest on the day you run your analysis, and report it with that date attached.

Further reading: Figshare vs Zenodo — Zenodo and Figshare compared: operator/funding, cost, file-size limits, licensing, curation, GitHub software archiving, institutional branding, and CoreTrustSeal certification status — with practical guidance on which to use for a given research data deposit.

For the wider identifier and bibliometrics landscape this guide sits inside — CRIS platforms, persistent identifiers, citation databases and discovery tools — see CASRAI’s CRIS and identifiers hub.