Author: Jason

Same topics, better assignments: a new topic classifier for OpenAlex

Posted on by Jason

Every work in OpenAlex now gets its topics from a new classifier. The topics themselves stay the same: same 4,516 topics, same names, same IDs, same 252 subfields, 26 fields and 4 domains. Your filters, dashboards and code keep working. What changes is which works are assigned to each topic, and the new assignments match what most readers expect: the topic fits what the title and abstract say.

It’s an incremental change. Institutions’ field profiles barely move, and university rankings by citation impact hardly change. Here’s what’s different, why, and what to do if you report on topics.

What’s different

Topics now match the title and abstract more often. Most people read a topic like a subject heading: they look at a paper’s title and abstract and expect the topic to describe them. The old topics often didn’t, and that confused people. One user wrote to us about a paper on universities: “it’s in ‘Diagnosis and Management of Kawasaki Disease’, though it’s about higher education in Australia.”

The mismatch came from how the old classifier worked. CWTS built the topics by clustering the citation network (our 2024 announcement), and the old classifier learned to predict which cluster a work belongs to, from its title, abstract, journal and references. For well-cited articles that works well, and a paper’s cluster usually matches its subject. Not always, and not always wrongly: a 2005 paper showing that hair follicle stem cells can repair severed nerves sat in “Hair Growth and Disorders”, because hair researchers cite it; the new classifier puts it in “Nerve injury and regeneration”. Both make sense. But for works with little to go on, the old classifier had to guess, and a few bugs made its guesses worse (below).

The new classifier reads each work’s title and abstract, in any language, and picks the topics whose subject matches. If the work has a venue, like a journal or a repository, it reads that name too, but it doesn’t need one, and it doesn’t use citations. So a dataset or a paper published yesterday gets topics as good as a well-cited article’s.

The gains are biggest where the old classifier had least to go on. We tested both classifiers on works that played no part in building the new one (how we tested):

Chart: share of works whose main topic matches the one two independent AI readers picked from the title and abstract, old classifier then new. Highly cited articles 70% to 89%. English articles and reviews with an abstract 48% to 88%. Works with no abstract 20% to 75%. Works not in English 9% to 72%. Datasets and other records that aren't papers 6% to 86%.

It keeps what the clusters mean. We also ran the test the old classifier was built for: publications CWTS clustered, held out from the old classifier’s training, each scored against the cluster CWTS put it in. The new classifier never saw those labels, yet it matches CWTS as often as the old one does, 55% each, and has CWTS’s topic in its top three more often (73% against 70%) (details).

The bugs it fixes

From biggest to smallest:

  1. One topic held 20 million unrelated works. The old classifier gave most papers in Japanese, Chinese, Russian and Korean with no abstract the same three topics with the same scores, as if it couldn’t see their titles at all. The first was “Military Technology and Strategies“: 20.6 million works, 1 in 18 works with a topic. Now it holds 169,000 works, and they’re about the military.
  2. Works with no abstract got weak guesses. Four in ten works have no abstract. The old classifier leaned on abstracts and references; with only a title, its main topic matched the work 20% of the time. The new one: 75%.
  3. Datasets and other records that aren’t papers got no topic, or a stray one. Three in ten works are datasets, specimen records, proposals and the like. The old pipeline scored only papers, so most of these got no topic, and a classifier trained on journal articles misread the rest: 6% matched. The new one: 86%.
  4. Papers not in English got weaker topics. A quarter of works aren’t in English. The old classifier learned mostly from English papers (83% of its training set); 9% matched, partly because of the first bug. The new one: 72%.
  5. Records with nothing to classify got a topic anyway. A table of contents or a bare file name now gets no topic instead of a guess.
Bug Work Old topic New topic
One topic for 20 million works Veselago’s 1967 paper on materials with negative ε and μ, in Russian, no abstract Military Technology and Strategies Metamaterials and Metasurfaces Applications
No abstract A paper on hand-gesture recognition with millimeter-wave radar Plant pathogens and resistance mechanisms Hand Gesture Recognition Systems
Not a paper A telescope proposal to map the tidal streams of globular clusters Methane Hydrates Galaxies: Formation, Evolution, Phenomena
Not in English A Spanish-language history of Yiddish-speaking communists in Buenos Aires Virology and Viral Diseases Argentine historical studies

Where the topics come from

CWTS at Leiden University built the topics in 2024 by clustering the citation network of OpenAlex works (our announcement). We built the classifier that assigns works to them. The classifier is what’s new; the clusters, their names and the hierarchy above them are untouched.

Your reports helped. Support tickets about topics showed us specific failure modes, like the 20-million-work bin, and the works in them became test cases. New AI models made the rest practical: a large model labeled two million works carefully, and we trained a small, fixed classifier on those labels. Its weights are open, so anyone can run it and get the same answer we do. The code, training labels and every test set are on GitHub.

If you report on topics

If you count your institution’s output by topic or field, or compare it with peers, here’s what changes and what to do.

Your field-level numbers barely move. We made before-and-after profiles for the 109 institutions that support OpenAlex. For the median one with at least 20,000 papers, its mix of fields is 97% the same and its mix of subfields 92% the same. Topics move more: 78% the same, and 7 of its top 10 topics stay in its top 10. Universities that publish a lot in languages other than English move most.

Rankings barely change. FWCI and citation percentiles compare a work with others in its subfield, so they shift when subfields shift. To see how much, we computed each university’s mean FWCI for its 2023–2025 papers both ways:

Scatterplot: mean FWCI of each of 3,621 universities' 2023–2025 papers, with the old topics against the new ones, on log scales. Nearly every dot sits on the no-change line, including the 166 universities that publish mostly in languages other than English.

Across 3,621 universities, mean FWCI with the new topics tracks the old closely: the ranking barely changes (rank correlation 0.99), and the median university moves 4.5%. Universities that publish mostly in other languages stay on the line too, moving about as much as English-language universities with similar citation impact. Why anyone moves: the old catch-all topics sat in a few subfields, like Aerospace Engineering, and their millions of rarely cited works dragged those subfields’ baselines down. With fairer baselines, universities strong in those subfields slip a little, and universities strong in languages and the humanities rise a little.

Topics now cover every kind of work, not just papers. The old pipeline gave topics to papers; the new classifier gives them to datasets, specimen records and every other kind of work too. That’s part of our commitment to open science. For nearly 20 years, studies have shown that sharing data pays off (2007, 2013), and reformers of research assessment, from DORA to CoARA, ask that datasets and software count as research outputs. When datasets have topics, they show up next to the papers in their field. A few of these collections are huge: about 14 million fusion-experiment shot records now sit, correctly, in “Magnetic confinement fusion research” (39 million counting the expansion corpus, which API queries leave out by default), and millions of specimen records in botany and fungi topics. If your report should count papers only, filter by type (articles, reviews, books and chapters), and those topics look normal again.

Check your own numbers. The 109 profiles are on GitHub. For any other institution, group its works by field on the website or the API (MIT since 2015, by field) and compare with your last report.

See what changed. The change-set files list every topic, subfield and field with its works before and after, where each old topic’s works went, and, for each topic, how often an independent judge agreed with its works before and after.

Need the old topics? They’re kept. Every work’s old topics and scores, frozen on 5 October, are attached to the v2.0.0 release (12 files, 9 GB). The old classifier stays public too: its code is on GitHub and its weights and training data are on Zenodo, so you can keep running it. Our text-tagging endpoint also keeps it, at /text/topics?version=1, until 13 January 2027, so you have three months to compare the two side by side. Snapshot users: this change doesn’t touch updated_date, so reload fully from the 14 October snapshot.

Also new: topics and keywords are connected

Each keyword now shows the topics it’s most used in, with the share of its works in each (“Related topics”), and each topic shows its characteristic keywords. A keyword can belong to several topics, and we show how much. Explore topics, their keywords and the whole hierarchy in the aboutness viewer.

A month of aboutness improvements

Topics are the last of a run of changes to how OpenAlex describes what works are about. In each, your reports showed us what to fix. The middle column is our accuracy on the old system, then on the new one; the notes below the table say how we measured each.

Change Old → new Links
Topics1 25% → 82% code
SDGs2 29% → 70% post, code
Keywords3 44% → 91% post, code
Study designs4 didn’t exist → 99.7% post, code
Semantic search5 43% → 55% list
  1. Topics: share of 2,000 random works whose main topic matches the one two AI judges (Claude Opus 5.5 and GPT-6.1 Sol) each chose blind from all 4,516, counted where the judges agreed (1,503 works). A strict test built on reading the text; with no AI involved, the field matches the category authors chose for their arXiv papers 75% → 83%. Details.
  2. SDGs: F1, which balances precision (29% → 68%) and recall (31% → 72%), on 2,000 random works, against the goals two AI judges (Claude Opus 5.5 and GPT-6.1 Sol) chose, with Fable 5.1 breaking ties.
  3. Keywords: share of keywords an AI judge (GPT-6 Astra) rated accurate, on 400 random works.
  4. Study designs: precision of the RCT filter, the share of works it returns that are RCTs, judged by Claude Opus 5.5 on 8,308 papers and weighted to the whole index. OpenAlex had no study-design field before. The filter finds 69% of the RCTs PubMed tags in MEDLINE.
  5. Semantic search: how close the first page of results comes to the best possible order (nDCG), on 657 real searches graded blind by an AI model (Fable 5.1).

Thanks to everyone who sends us examples when something looks off. Keep them coming.

OpenAlex’s SDG tags are now right 7 times in 10, up from 3

Posted on by Jason

OpenAlex has a new classifier for the UN Sustainable Development Goals, and it’s much more accurate than the old one. On 2,000 works drawn at random from OpenAlex, its SDG tags are right 68% of the time; the old model’s were right 29% of the time. It replaces the old tags on every work today.

BenchmarkNew classifierOld classifier (Aurora SDG-BERT)
2,000 random OpenAlex works, judged by AI judges (F1)0.700.29
…precision (share of tags that are right)0.680.29
…recall (share of right answers found)0.720.31
Goals where it does better (of 17)170
8,767 researcher votes from the Aurora survey (F1)75.563.6

What was wrong with the old tags. Our SDG tags came from Aurora’s SDG-BERT model, which we’ve served for years. On a random sample of OpenAlex, about 7 of every 10 of its tags were wrong. It was worst on energy: SDG 7 was its single biggest label, and only about 1 in 20 of those tags held up. Several of you told us so, and you were right.

How we measured it. We drew 2,000 works at random from all of OpenAlex, including the 37% that have only a title. Two AI judges from different labs (Claude Opus 5.5 and GPT-6.1 Sol) read each work against the UN’s own text for each goal and its targets, independently; where they disagreed, a third model (Claude Fable 5.1) decided. The judges agreed with each other on 98.9% of decisions. The ranking held under every judge on its own. We also scored both models on the 8,767 paper-and-goal questions from the Aurora survey, where 244 researchers voted on whether papers belong to a goal; these papers came from Aurora’s own search queries, so it’s the old model’s home turf. The new classifier wins there too.

How it works. A large language model read titles and abstracts for 200,000 works and decided, goal by goal, whether each work addresses that goal. We trained a small classifier on those decisions, using the vector OpenAlex already stores for every work. That makes it cheap to run on all 475 million works, and new works get tagged within a day or two of being added. The training labels, the 2,000 judged works, the judging rubric and the code are on GitHub: ourresearch/openalex-sdgs.

What counts as an SDG. A work counts when its main subject addresses one of the goal’s UN targets. Being about the topic isn’t enough: a paper on atomic energy levels isn’t SDG 7, and a species description or a museum specimen record isn’t SDG 15 (life on land) unless it’s aimed at conservation or land use. Researchers who answered the Aurora survey were split on pure taxonomy papers; we went with the stricter reading, because the people who use this field most are counting research that contributes to the goals.

What you’ll see. About a third of works now carry at least one goal. Counts per goal move a lot:

GoalOld tagsNew tags
3 Good health and well-being26.8M75.4M
16 Peace, justice, and strong institutions13.0M18.8M
4 Quality education18.0M18.5M
2 Zero hunger19.0M15.9M
9 Industry, innovation and infrastructure6.5M12.4M
8 Decent work and economic growth11.4M7.8M
10 Reduced inequalities14.2M6.1M
11 Sustainable cities and communities11.4M6.1M
7 Affordable and clean energy25.1M5.9M
13 Climate action6.4M5.4M
12 Responsible consumption and production1.5M4.7M
15 Life on land8.1M4.6M
6 Clean water and sanitation7.7M4.1M
5 Gender equality9.6M2.8M
14 Life below water9.7M2.3M
1 No poverty2.1M1.6M
17 Partnerships for the goals3.7M0.9M

If you report SDG counts for your institution, your numbers will change, probably by a lot; the new ones are much closer to right. We’d love to hear where they still look off: support@openalex.org.

For API and snapshot users:

  • sustainable_development_goals keeps the same shape: id, display_name, score.
  • score is the classifier’s confidence from 0 to 1, and a goal appears at 0.4 or above. Higher means surer, but it’s not a probability: in our checks, tags scored just over 0.4 were right about a quarter of the time, 0.5 to 0.8 about half the time, and above 0.9 about 8 times in 10.
  • updated_date did not change for this switch. If you sync incrementally, reload this field from the snapshot.
  • The old tags stay available for about a month in a deprecated field, sustainable_development_goals_aurora (read-only), then go away in early November.
  • The experimental x_sdgs filter is gone.

Details, per-goal numbers and how to read the scores: help.openalex.org/data/sdgs.

Jason

OpenAlex keywords are finally good

OpenAlex keywords used to be bad. Now they’re good. 419 million works in OpenAlex have new keywords, drawn from a vocabulary of 1.88 million. They’re 91% accurate, up from 44% for the old keywords.

The old keywords were guessed from each work’s topics, so they were often off. The paper announcing the first gravitational waves got “Black hole (networking)” and “Binary number.” The new tagger reads each work’s title and abstract and says what it’s about:

PaperOld keywordsNew keywords
Observation of gravitational waves from a binary black hole merger (2016, here)Black hole (networking), Binary number, Computer sciencebinary black hole mergers, LIGO, ringdown, tests of general relativity
Deep learning (2015, here)Representation (politics), Layer (electronics)backpropagation, convolutional neural networks, recurrent neural networks
Highly accurate protein structure prediction with AlphaFold (2021, here)Computer science, Function (biology)AlphaFold, protein structure prediction, protein folding problem, multi-sequence alignment

How much better

Share of keywords an independent judge rated accurate, old vs new. All works 44% to 91%; with an abstract 47% to 94%; no abstract 29% to 78%; not in English 19% to 84%.

Our judge is an AI model from another lab: OpenAI’s GPT-6 Astra, since our keywords were built with Anthropic’s Claude. It read 400 random works and rated each keyword on its own. You can see the results above. Shown both lists side by side, the judge preferred the new keywords for 395 of the 400 works.

The new keywords also match the keywords authors chose for their own papers twice as often as the old ones did, and they beat our previous best models on two standard keyphrase benchmarks. Every number is here.

The new keywords cover 89% of works, down from 95%. That’s on purpose: when a record doesn’t say what it’s about (a bare dataset entry, a table of contents), the tagger now returns nothing rather than a guess.

What they’re good for

  • Finding what text search misses. A keyword finds the paper even when authors used different words for the same concepts. 133 million works with no abstract still get keywords.
  • Mapping a field. Group any set of works by keyword to see what it’s made of, and by year to see what’s rising.
  • Finding experts. Filter to a keyword and group by author; in our tests the top ten were real specialists on 24 of 24 queries.
  • Searching across languages. Keywords are in English, but papers in any language get them: a Chinese paper on microplastics in soil carries the same microplastics keyword as an English one. About half of what the keywords add to a search on microplastics or remote work isn’t in English.
Keywords per paper as a share of what an English-language paper gets, old vs new: Portuguese 39% to 106%, Indonesian 40% to 104%, Turkish 35% to 104%, Spanish 39% to 96%, Arabic 24% to 94%, Korean 51% to 94%, Italian 37% to 90%, German 31% to 86%, Russian 48% to 84%, French 31% to 82%, Japanese 27% to 81%, Chinese 16% to 74%.

How to: find the papers your text search misses

Say you’re studying antimicrobial resistance. Just search: OpenAlex search now reads keywords along with titles and abstracts. When a phrase you type names a keyword, a paper matches if the phrase is in its title or abstract or the paper carries that keyword. A title and abstract search for the phrase finds 118,920 works; this one finds 241,167:

https://api.openalex.org/works?search.title_abstract_keywords="antimicrobial resistance"

The antimicrobial-resistance keyword brings in about 122,000 of them, and our judge rates 81% of those on topic: about 100,000 relevant works the phrase search missed, so you find about twice as many. Some say “antibiotic resistance” or “drug-resistant bacteria” instead. 47,269 have no abstract at all, and 74% of those are on topic, like “Increased resistance to antibiotics among microorganisms isolated in a general hospital” (1958, here) and “Lack of development of new antimicrobial drugs” (2005, here); no text search could find those.

On openalex.org there’s nothing to set: “Title, abstract, & keywords” is now the search box’s default, and the top results are reranked so the papers most clearly about your topic come first. Every other word you type still has to be in the text, so “remote work productivity” finds papers tagged remote work that mention productivity, never a paper tagged only productivity. In OQL it’s works where title-abstract-keywords has ("antimicrobial resistance").

Let your AI agent do a thorough search

The search catches the keywords your own words name. An AI agent can go further: find the keywords that mean your topic in other words, add them, and read the results to drop the strays. The /text/keywords endpoint takes any text, including a title or abstract, and returns the keywords that fit it, and an agent like Claude Code or Codex can use it for you. Tell it something like this:

Check help.openalex.org first, then use the OpenAlex API to find papers on <your topic>. Use the title, abstract and keywords search, and add OpenAlex keywords that describe my topic in other words, all in one query. Each paper should cover every part of my topic. Read the results as you go and keep only the papers that are really about my topic. Save the 200 most-cited of those as a CSV and show me the query and how many papers each part found.
My API key: <your key>

Keep the “check help.openalex.org” part: without it, agents work from what they remember about the OpenAlex API, which is out of date. We tested this prompt on fresh installs of Claude Code and Codex: every run found the title, abstract and keywords search on the help site and built one query with it, and on remote work and employee wellbeing, 99% of the papers in their final lists were at least partly on topic (about seven in ten clearly so). In the Claude app, set the chat to Auto (next to the model name, under the message box), or it will ask your permission before every search. You’ll need a free API key; the full recipe is here.

Or, if you use Claude, just ask. Add the free OpenAlex connector, now in Claude’s connector directory, and ask in one sentence: “Do a thorough search for open-access papers on how remote work affects employee wellbeing.” The connector searches titles, abstracts and keywords, adds keywords that mean your topic in other words, reranks the results so the most relevant come first, tells you how many works each part found, and hands back the query so you can rerun it.

The vocabulary keeps changing

This isn’t a fixed list. New fields show up every year, and we’ll add keywords for them as they do. We already merge keywords that mean the same thing (“Neanderthals” and “Neandertals”), and we’ll keep merging, splitting and adding as you tell us what’s wrong. Send us feedback on any keyword that misses.

Keywords are already linked to Wikidata: about 22% of keywords have a Wikidata id, and those cover about 56% of keyword uses. That makes it easy to join OpenAlex keywords to other data. Keyword search also knows 5 million synonyms, so searching keywords for “heart attack” finds myocardial infarction. Next, we’ll look at joining keywords to other outside vocabularies like MeSH, and at vector search, so that searching the keywords for “antibacterial resistance” finds antimicrobial resistance. We also want to assign each keyword to one or more of our topics, so keywords become part of the whole OpenAlex aboutness taxonomy.

Check our work

Everything is open: the code, prompts, benchmarks, vocabulary and model weights, with a release that lets you rerun the tagger yourself. The keywords help page has the details.

Edit, 3 October 2026: An earlier version of this post showed how to run a title and abstract search and a keyword search side by side and combine them by hand. A few hours after it went out, we shipped a search that does both at once (title, abstract and keywords, now the default on openalex.org), so we rewrote the how-to section and the AI prompt to use it.

Edit, 3 October 2026, later the same day: The OQL field in this post is now spelled title-abstract-keywords (it was title/abstract/keywords), because slashes already separate the parts of OpenAlex IDs. The old spelling still works.

Filter OpenAlex for RCTs, meta-analyses and systematic reviews

You can now filter all of OpenAlex by study design: randomized controlled trials, systematic reviews, meta-analyses and more. 21 million works, in every field and language, carry one. Our randomized-trial tags are 99.7% precise, and on the same papers they’re more precise than PubMed’s, long the gold standard.

When you’re searching, you can focus on just the highest-quality evidence. And if you build evidence tools, you get study design for the whole literature, free, through the OpenAlex API.

Seven designs, every field

The designs come from PubMed’s own vocabulary, and we add nothing to it. What’s new is reach. PubMed tags the journals it indexes; we tag every work with an abstract. Half of the randomized trials we tag, 306,000, are in journals PubMed’s MEDLINE doesn’t index.

Here’s a graph how many works are tagged with each study design, in both OpenAlex and PubMed:

Works tagged with each design, OpenAlex then PubMed: observational study 14,480,358 and 217,835; case report 3,160,292 and 2,241,251; clinical trial 1,864,938 and 1,068,573; systematic review 1,146,957 and 415,656; randomized controlled trial 622,899 and 677,175; meta analysis 384,599 and 243,120; study protocol 106,024 and 24,147

Every randomized trial is also a clinical trial, and every meta-analysis is also a systematic review, so a work can count twice. PubMed’s Observational Study tag dates only from 2014. Works without an abstract get no design.

Find the trials, see the evidence

Filter to one design. Of 34,395 papers that mention semaglutide, 533 are randomized trials; one filter returns just those, on openalex.org or in the API with filter=study_designs.id:randomized-controlled-trial. AI agents that search OpenAlex can do the same.

Group by design to see what kind of evidence a field has. Long COVID research is mostly observational: 31 observational studies for every randomized trial. Semaglutide already has more meta-analyses (1,341) than randomized trials.

Tagged papers on each topic, by design. Papers an OpenAlex search for the topic returns.

Share of each topic's tagged papers by design: trials, observational studies, case reports, and systematic reviews and meta-analyses, for intermittent fasting, mindfulness, semaglutide, vitamin D and long COVID.

Share of each topic’s papers that carry a design; protocols left out. We tag conservatively, so real counts run higher, trials most of all.

Built on PubMed, tuned for precision

PubMed’s publication types are carefully indexed and trusted across medicine. But its tags sometimes describe the study a paper came from rather than the paper itself, so a methods paper or a later analysis of trial data can carry the trial’s tag. We tag only what the paper itself reports, and only when we’re confident.

How often each tag is right

Share of tags an independent judge confirms, on the same PubMed-indexed papers.

Share of tags an independent judge confirms, PubMed then OpenAlex: randomized controlled trial 63% and 99.9%; clinical trial 65% and 96.5%; observational study 89% and 97.6%; case report 89% and 98.8%; systematic review 91% and 98.0%; meta analysis 88% and 98.9%; study protocol 99% and 99.6%

Where the tags differ

Here are some examples of papers that PubMed tags as RCTs, but we (correctly) don’t apply the RCT.

PaperWhat it is
Bland and Altman, Lancet 1986: measuring agreementA statistics methods paper
TOAST stroke classification, 1993Definitions written for a trial
The Implicit Association Test, 1998Lab experiments that built a test
Oncotype DX gene test, NEJM 2004A gene test on tumors from one arm of an earlier trial
SAPS II severity score, JAMA 1993A risk score built from a cohort

Our model is tuned for precision, not recall–when we say it’s an RCT, it’s an RCT, with 99.9% accuracy. PubMed captures some RCTs that our stingy approach misses, but at the cost of some accuracy.

Judge: Claude Opus 5.5, under written definitions, on 8,308 papers sampled so that every estimate is weighted to the whole index; nothing was tuned on that sample. Re-reading 436 open-access papers in full barely moved the verdicts.

Check our work

Everything behind these numbers is public: the code, the models, the samples and every judge’s verdict. One command reruns every table. github.com/ourresearch/openalex-study-designs

New Features and Usage-Based Pricing

Today we’re adding some features to the OpenAlex API: better search, content download, and new docs. Most importantly, we’re also introducing usage-based pricing.

New features

Advanced search at last

We’ve had lots of request for advanced search features to support systematic reviews. Good news: they’re here!

  • Proximity search: find terms near each other
  • Exact matching: skip stemming when you need precision
  • Wildcards: for when you’re not sure of the exact form
  • Lonnnnng queries: Searches can be up to several pages in length (8kb)

Find details and examples of advanced search in the new developer docs here.

Note for developers: the old filter syntax for search is now deprecated; the ?search= parameter approach remains. It’ll be the One Way To Do It moving forward. Filter searches will redirect to the ?search param.

Semantic search

We’re also launching semantic search. Instead of just matching keywords, it uses embeddings to match the meaning of your search–so a search for “kelp biomechanics” also finds articles about algae and wave mechanics. But you don’t have to stop there: you can even paste a whole abstract into the search bar to find related papers!

Semantic search is in beta; we don’t recommend using it for sensitive production workflows yet. But we would love to hear your feedback! If it’s well-used we’ll continue to invest more resources into it.

Full-text downloads

We’re hosting PDFs and TEI XML for our 60M open-access works. You can search and filter for works of interest, filter to get just ones with PDFs, and then download the PDFs in bulk—all with the API. Or you can use our new OpenAlex CLI to do it from the command line, massively parallelized, in a single command. Or your agent can—they love CLIs.

openalex download \ 
  --api-key YOUR_KEY \ 
  --output ./climate-pdfs \ 
  --filter "topics.id:T10325,has_content.pdf:true" \ 
  --content pdf

See the full-text documentation for details.

New docs

We’ve completely rebuilt our documentation. The old docs are deprecated and will redirect soon. The new docs are clearer, cleaner, up to date, and AI-optimized. We want to make OpenAlex as easy as possible to use for everyone, whether they’re an expert or a novice vibe-coding their first app.

API keys are now required.

As we announced in January, you’ll need an API key for all requests. Getting one is free and takes about 30 seconds: create an account at openalex.org, then grab your key at openalex.org/settings/api. You can still make a few calls without an API key for demo purposes, but it’s not suitable for any kind of production use. The API keys are essential for our new usage-based pricing model. What’s a usage-based pricing model? Gentle Reader, a mere centimeter now separates you from the answer. 👇

Usage-based pricing

Different API operations cost us different amounts to run. Doing stuff with PDFs is expensive, but looking up a single work by ID is nearly free. We think it’s essential that our pricing reflects these actual costs. Usage-based pricing is a natural fit for this: it’s transparent, sustainable, and fair.

Here’s what things cost. See the developer docs for more details.

endpointcost per callcost per 1,000 calls
single work lookup by DOI or ID00
list and filter$0.0001$0.10
search$0.001$1.00
PDF/XML download$0.01$10.00

Free usage

Every API key gets $1 of free usage per day. We’ve always subsidized free users using revenue from paying ones–this makes the exact extent of that subsidy clear, transparent, and unambiguous.

What does that daily dollar get you? Assuming you return 100 works per request:

endpointdaily free callsdaily free results
single work lookup by DOI or IDunlimited unlimited
list and filter10,0001,000,000
search1,000100,000
PDF/XML download100100

To use a real-world example: grabbing all 694k works by Finnish authors takes about 7k paginated requests at $0.10 per thousand or $0.70. That’s covered by your free daily allowance. But if you want all 9 million works from Japan, that will cost about $9. (You could even download all 480M works in OpenAlex this way for $480—but don’t do that lol, download the full dataset instead, it’s free).

It’s easy to track your usage: every API response includes headers showing how much you’ve spent and how much you’ve got left. You can also check openalex.org/settings/usage anytime.

Prepaid usage

Most users will find that the free plan covers all their needs. However, for some projects, you may need more usage. The great thing about usage-based pricing is that most of the time this will only cost you a few bucks. You’re just paying for what you need. You can buy prepaid usage in 1min with your credit card, whenever you want, however much you want. It supplements your daily free allowance.

Organizational plans

Organizations also buy prepaid usage. But many will want to get annual plans instead, which offer major discounts, data sync, curation dashboards, and more. Check out our new Member, Member+, and Supporter plans for more details.

FAQ

I thought it was free? The data remains free. The full OpenAlex dataset—all 480M works, all the metadata—is free to download, share, remix, and build on. We’re committed to keeping it sustainably free by charging for a service (the API) built on that dataset. Free data, paid service–this is the path laid out in the POSI principles, which we’ve signed and enthusiastically support. 

How do I track my usage? Every API response includes usage info; you can also call the rate-limit endpoint or check your usage page on openalex.org. Learn more here. 

How is my usage data used? We analyze usage data to improve the overall service and we provide institutions aggregated usage summaries for their institutions upon request. We only collect what we need to run OpenAlex. We aren’t building tools to monitor individuals and we don’t sell your data. You can read our full privacy policy here.

Why charge per request instead of per result? We’re trying to link our costs to our pricing, and our costs mostly scale with requests, not results; a search that returns 10 results costs us about the same as one that returns 10,000.

Will prices change?

Yes, probably. The point of this model is to keep our prices tightly linked to our costs, and our costs will likely change with new tech, new use cases, and new data.

Where from here?

AI accelerates every day. The future of knowledge is getting rebuilt, right now. If we build on checkerboards of enclosed, walled gardens, we build a fragmented, incoherent future for scholarship and humanity.

We think OpenAlex can help with that. We’re gathering and connecting the literature into a cohesive living library, complete and organized and accessible to everyone. Today’s new pricing model helps us stay in this for the long haul.

An API-based sustainability model lets us deliver (and monetize) value in the post-GUI era. Soon, users won’t go to openalex.org (or any SaaS website), they’ll use APIs to vibe-code a custom interface for any question in minutes. [1] The post-GUI world will be tough on some open sustainability models. But it’s also an amazing opportunity for open infrastructure, if we adapt our pricing model correctly. That’s what we’re doing today.

We’re so very excited about this next chapter. Questions? Hit us up at support@openalex.org.

Let’s build!

[1] Check out our Q1 town hall for more on our post-GUI strategy, and check out this vibe-coding webinar to see several real-life examples of building five-minute custom OpenAlex dashboards.

OpenAlex 2026 Roadmap

We just wrapped up our Q1 2026 Town Hall. You can watch the full recording here, but this post covers the highlights: what we shipped last quarter, what’s coming this quarter, and why we think 2026 is a pivotal year for open science.

What we shipped in Q4

The Walden rewrite is done. OpenAlex now runs on a modern Databricks infrastructure that lets us ship faster and iterate on data quality in days instead of months.

We added 192 million new works from DataCite and repositories. OpenAlex now indexes 477 million works—the largest connected repository of scholarship ever published.

On funders and awards: we created Awards as a first-class entity, extracted 27 million funder links from fulltext PDFs, and integrated 15 new funders directly.

What’s coming in Q1

For enterprise users: Credit-based API pricing launches this month. Different calls cost different amounts:

  • a singleton (/works/w123) is 1 credit,
  • a list (/works?filter=foo:bar) is 10,
  • PDF content (coming this month!) is 100,
  • vector search is 1,000. (coming soon! email steve@ourresearch.org for early access!)

We’re also launching a sync service so you can pull daily updates in one chunk instead of polling millions of records.

For institutions: Affiliation matching curation launches in February. Members can edit the matching algorithm that links affiliation strings to their institution. Changes propagate to the API within a day—permanently improving the dataset for everyone.

We’re also launching two membership tiers at $5k and $20k/year that include ability to curate your own data in OpenAlex, training/consulting, and pro API keys with higher API access for your faculty.

For researchers: A complete rewrite of author name disambiguation ships by end of Q1. This has always been the hardest problem in bibliometrics. With today’s AI, we think we can build the most accurate system ever made.

The bigger picture

There’s a lot more I want to say about why 2026 feels like a pivotal year—why we think the GUI is dead, why open data wins the AI era, and what that means for OpenAlex. I’ll save that for a follow-up post. For now: watch the town hall to hear the full argument, and try the vibe-coded demo I built live during the talk. And join our mailing list to stay up-to-date on all the wild stuff we’re doing this year. It’s going to be, by far, our biggest year ever. You ain’t seen nothing yet.

OpenAlex rewrite (“Walden”) launch!

Today, OpenAlex gets a new engine.

After a year of rebuilding, refactoring, and retesting, the Walden rewrite is now live — powering all of OpenAlex. It’s the same dataset shape you know, but faster, cleaner, and more complete.

You’ll notice better references, better OA detection, better language and license coverage, better everything. We’ve added 190 million new works, including datasets, software, and other research objects from DataCite and thousands of repositories. And thanks to our new foundation, fixes and improvements now roll out in days, not months.

Want to see exactly what changed? Check out OREO — the OpenAlex Rewrite Evaluation Overview — to compare old vs. new data in detail. [edit Dec 13, 2025: OREO is no longer up because the legacy OpenAlex data is no longer being updated…it’s all Walden now, so there’s no comparator].

And if you’d like to dig into the full list of updates, the Walden release notes have you covered.

For the next few weeks, you can still access the old dataset with data-version=1, and starting tomorrow, you can download full snapshots of both the legacy and Walden datasets in the usual way.

The rebuild is done. The road ahead is wide open.

Onward.

OpenAlex rewrite enters beta! 🎉

It’s a big week at OpenAlex. On Monday, we announced that OpenAlex is now our top-level brand (and retired the “OurResearch” name). Yesterday we unveiled our new logo. And today, we’re thrilled to launch the beta release of our fully-rewritten codebase (codenamed Walden)!

Walden is faster, bigger, and more maintainable–that means quicker bug fixes, more content, easier feature development, and a smoother experience all around.

Throughout October, we’ll be running Walden and the old system (Classic) side by side, with Classic remaining the default. On November 1 2025, Walden becomes default, and we’ll publish the last data snapshot from the old system (more info on timelines here).

How to test-drive Walden

Walden beta is already live in the API and UI so you can start exploring it right away!

Just remember that it’s still in beta: there are lots of known issues and it’s changing every day. If you notice an that’s not already in OREO tests or known issues, report it here.

Key improvements

When you check it out, what should you expect to see? The best way to view a list of improvements is to check out the tests in OREO, especially work tests. But here’s a high-level overview:

  • 150M+ new works: Newly indexed articles, books, datasets, software, dissertations, and more! You can explore just the newly added works here.
  • Better consistency: Unpaywall and OpenAlex will now always agree.
  • Better metadata: more citations, more language and retraction coverage, better keywords, more OA data.

Looking Ahead

The last year of rewriting OpenAlex was tough. We couldn’t move as fast as we wanted on new features, and support often lagged. But now we’re equipped to move fast without breaking things. Expect faster improvements, better support, and more ambitious features dropping in Q4, including:

  • Community curation: fix mistakes (like in Wikipedia) and see them reflected in days.
  • Vector search endpoint: find relevant works and other entities based on semantic similarity of free-form text
  • Download endpoint: Access PDF text from DOI or OpenAlex ID
  • Better funding metadata: New grants entity with better coverage of grant objects and linkages to research outputs and funders

This is a turning point for OpenAlex—and we’re excited to build the future of research infrastructure together with you. The engine’s rebuilt. The road ahead is wide open. Let’s go.

PS want to learn more about Walden? Come to our webinar Oct 7th at 10am Eastern. You can register to attend here.

A New Logo for OpenAlex

This is a big week for OpenAlex: yesterday we reorganized under the OpenAlex brand, and tomorrow we launch our completely rewritten codebase (beta). Today we launch our new logo!

The old logo was unique and conveyed the idea of building, which we loved. But was also visually complex, almost Escheresque; consequently, it didn’t scale down well, and it failed to convey the directness, boldness, and simplicity of our vision: to create a universal, open library of scholarly information.

So as we start a new chapter in OpenAlex, it’s a great time to also launch a new look.

You Bring the Color

We’re doubling down on black and white. That’s not just a design choice, it’s a statement of philosophy. OpenAlex is infrastructure. We’re the pipes under the city, not the flashy towers above it. We want to stay out of your way and let your projects, your creativity, your insights provide the color. You’ll see this commitment carry through in our website, which now leans harder into that clean, monochrome look.

A New Typeface

We’ve switched to Inter. Of course it’s open, just like us. Inter is modern and businesslike, but still human, readable, and approachable. Compared to Dosis (our old font), Inter is sharper, more confident, and more professional—while keeping the sense of openness that’s core to who we are. You’ll see Inter across our site from now on.

The Icon

The new icon is simple: a single, continuously curved outline forms three joined dots. Individually, dots are just dots—but when you connect them, something new emerges:

  • It’s an A for Alex—but sans crossbar, offering an open doorway in.
  • It’s a connected network—could be works and citations, coauthorships, or any of the billions of nodes and edges in the OpenAlex graph.
  • It’s a simplified water molecule.

Why Water?

Water has increasingly become part of our story of what OpenAlex is. Water’s simple but essential. We count on shared infrastructure to deliver it, quietly and reliably, cheaply, but not for free.

At OpenAlex we want to be the pipes under the scholarly city: infrastructure that delivers research information wherever it’s needed, at scale, for cheap. We’re here to support all the amazing things the research ecosystem is doing—quietly, reliably, and everywhere.

A Modern Library of Alexandria

The original Library of Alexandria aimed to collect all scholarly knowledge. OpenAlex means to carry that spirit forward in the digital age: building an open, connected, comprehensive graph of the world’s research.

Our new logo—an open A, a network, a molecule of water—is our way of putting that vision front and center. And if you see a pyramid in this logo, it’s not an accident, it’s homage.

The new logo is a reminder of what we’re building: a simple, open, essential infrastructure for the world’s research information: cheap, reliable, everywhere. That’s OpenAlex.

PS for logo nerds, other inspo: Vercel, the Banner of Peace, and iconic Paul Rand banger Westinghouse.

We’re now OpenAlex

For years, we’ve been working under the name OurResearch. That name sat at the top of our org chart, with three child projects under it: OpenAlex, Unpaywall, and Unsub.

Starting today, things are simpler: that org chart now has just one parent—OpenAlex—with Unpaywall and Unsub beneath it.

Why the change? Three reasons:

1. Fewer brands is clearer

We’re a tiny team, and having so many brands has always been confusing. People wondered: are we OurResearch (or Our Research), or OpenAlex, or Unpaywall, or something else? From now on, the answer is simple: we’re OpenAlex.

2. OpenAlex is what we do

More and more, OpenAlex is the center of our work. It’s our biggest project and the one that takes most of our time. And it’s also the data engine behind our other projects: Unpaywall and Unsub both run on OpenAlex data. In fact, with the launch of our fully rewritten OpenAlex codebase (codenamed Walden) this week, Unpaywall runs as a subroutine of the OpenAlex codebase.

So in a real sense, Unpaywall and Unsub are just friendly wrappers around OpenAlex. Improving OpenAlex improves them automatically.

And the name OpenAlex, with its homage to the ancient Library of Alexandria, captures our long-term vision to gather, organize, and make open all scholarly information.

3. New name, new start

Legally, nothing dramatic is happening—our official name has always been Impactstory, Inc., and “OurResearch” was just a DBA. But this moment is more than just a bookkeeping change.

This is a new chapter for us. The past year has been tough: not much visible progress, a lot of repaying technical debt, and a long slog to rewrite our entire codebase. But that rewrite launches (in beta) this week. And with a fresh codebase comes a fresh start: we get to focus harder, move faster, and pour our energy into making OpenAlex as comprehensive, accurate, and open as possible.

So yes, the name change simplifies things. But more importantly, it marks a new focus and a renewed commitment to our vision: building a universal library of scholarship.

And while we’ll continue to support Unpaywall and Unsub for now, we want to be transparent: OpenAlex is the future. As its functionality grows over the next year or two, Unpaywall and Unsub users will be able to meet their use-cases directly via OpenAlex. The rising tide of OpenAlex lifts all boats.

This week is about OpenAlex

This post is the first of three announcements:

  • Monday: our name change to OpenAlex (that’s today).
  • Tuesday: our new logo.
  • Wednesday: the beta launch our fully rewritten OpenAlex codebase.

When we say we’re focusing on OpenAlex, it’s not just words—we’re shipping, this week. And there’s more coming in Q4:

  • A new API endpoint to directly download PDFs and parsed PDFs.
  • A self-serve curation portal (think Wikipedia editing, but for scholarly metadata), where your changes go live in a day or two.
  • A new vector search API.
  • Improved funder coverage, thanks to our new Wellcome Trust grant.

After a year of rebuilding, we’ve finally got the tools and the focus we need start delivering more substantively on our vision: a universal, open library of scholarly information. We’re energized. We’re ready. We’re OpenAlex.