Tool profile
Semantic Scholar: cost, fit and limits
Semantic Scholar is a free academic search and data service run by the Allen Institute for Artificial Intelligence (Ai2) in the United States, with an Academic Graph API covering papers, authors and citations. Its data feeds many other research tools.
What to know first
Semantic Scholar has no paid plans: most API endpoints work without a key, and a key with higher limits is requested by form and delivered by email. The API page lists about 214 million papers and 2.49 billion citations, plus SPECTER2 embeddings and a similar-paper recommendation service. The pages opened describe an API and datasets, not a generative question-answering assistant, so it fits builders more than readers wanting written answers.
Semantic Scholar at a glance
| Maker | The Allen Institute for Artificial Intelligence (Ai2), United States |
|---|---|
| Free plan | Yes |
| Entry price | Free |
| API access | Yes |
| Category | Best AI Literature Search Tools in 2026, rank 8 of 13 |
| Website | semanticscholar.org |
Who Semantic Scholar is best for
Developers building research tools
The Academic Graph API returns papers, authors, citations and venues, and most endpoints work without authentication, so a prototype needs no payment.
Data scientists studying citations
The Datasets service offers downloadable links for S2AG, SUPP.AI and S2ORC, which suits bulk analysis that rate-limited search cannot support.
Tool makers needing similarity search
SPECTER2 embeddings and a recommendations service return similar papers, which can power a related-work feature in another product. The API page offers it as a building block.
Budget-limited labs
There is no price on any listed service, so a group can build on the data without a procurement process.
Semantic Scholar plans and pricing
| Plan | Price | What it includes |
|---|---|---|
| Free, no key | Free | Most endpoints without authentication; unauthenticated users share a limit of 1,000 requests per second, with throttling during heavy use. |
| API key | Not published, requested by form | Higher rate limits with an introductory limit of 1 request per second on all endpoints; the key is delivered by email. |
| Datasets | Not published on the API page | Downloadable dataset links for S2AG, SUPP.AI and S2ORC through the Datasets service, for working offline on the corpus. |
Checked November 2026.
What Semantic Scholar does
- Academic Graph API
- The API covers papers, authors, citations and venues. The page lists about 214 million papers and 2.49 billion citations, which are the vendor's own figures.
- SPECTER2 embeddings
- The Academic Graph service includes SPECTER2 embeddings, which represent papers as vectors so similar papers can be found by distance rather than by shared keywords.
- Recommendations service
- Given a set of papers, the service returns similar-paper suggestions, which developers can embed as a related-work panel in their own application. Applications can show it as a related-work panel.
- Datasets service
- Download links cover S2AG, SUPP.AI and S2ORC, which suits researchers who want to run their own analysis on the corpus instead of querying page by page.
- Open access to endpoints
- Most endpoints need no authentication, so exploring the data takes minutes and needs no account or billing details. No account or billing details are needed to start testing queries against the service.
How Semantic Scholar handles your data
The Ai2 privacy policy does not mention training AI models, but it says prompts entered into Ai2 AI demonstrations may be reviewed to improve the services. Data is kept only as long as necessary for the purposes collected, with no periods stated except 14 months for Google Analytics data. Personal data is stored and processed by Ai2 in the United States, and transfers rely on Standard Contractual Clauses and Data Privacy Frameworks. The policy claims no certifications. Queries to the public API are not covered by a separate data policy on the page opened.
Where Semantic Scholar falls short in practice
- Slow key limit. An introductory API key allows 1 request per second on all endpoints, which makes bulk harvesting slow without a request for a higher limit. Ask for a higher limit when planning bulk use.
- Shared public limit. Unauthenticated users share a pool of 1,000 requests per second that is throttled in heavy use, so a script can slow without warning. A burst of requests can fail when other users load the service.
- No assistant described. The API page names embeddings and recommendations but no generative assistant, so readers wanting cited written answers must look elsewhere. Anyone expecting summaries from a chat window will not find them here.
- No support terms. No paid tier is listed, so no support terms, service levels or enterprise commitments are published on the pages opened. Organizations that need a contract should ask Ai2 directly.
Works with
Alternatives to Semantic Scholar
Semantic Scholar: common questions
Is the Semantic Scholar API free?
Yes. The API page lists no prices, most endpoints work without authentication, and an API key with higher rate limits is requested by form and emailed. An introductory key allows 1 request per second on all endpoints. Ask for a higher limit if your project needs more.
How large is the Semantic Scholar corpus?
The API page lists about 214 million papers and 2.49 billion citations. Those are the vendor's own figures from the page opened in November 2026, and coverage varies by field, so test your own subject before relying on the numbers.
Does Semantic Scholar use AI to answer questions?
The API page names SPECTER2 embeddings and a recommendation service for similar papers. It does not name a generative question-answering assistant. The Ai2 policy mentions prompts entered into its AI demonstrations, which are a separate product area. Check which product you are reading about before assuming chat exists.
Where does Semantic Scholar store data?
The Ai2 privacy policy says personal data is stored and processed by Ai2 in the United States, with transfers resting on Standard Contractual Clauses and Data Privacy Frameworks. No certification is claimed in the policy. Ask Ai2 about regional options if your funder requires EU storage.
Can I download the whole Semantic Scholar dataset?
The Datasets service provides downloadable links for S2AG, SUPP.AI and S2ORC. The API page does not list file sizes or license terms in the text we read, so check the license before redistributing any data. Confirm the terms of each dataset before building a product on it.