Crossref、OpenAlex、CORE 与 Unpaywall：学术元数据 API 对比

Choosing an academic metadata api starts with your product's output. Does the user need a DOI record, a downloadable paper, or an evidence-backed comparison?

Crossref, OpenAlex, CORE, and Unpaywall provide different retrieval building blocks. SciSpace addresses the research workflow after and around retrieval: searching, reading, extracting, and writing cited reports delivered as files. The comparisons below are architectural recommendations, not measured quality rankings.

Which academic metadata API should you choose?

Shortlist by required output rather than corpus size. These interfaces are not interchangeable.

API	Best starting point	Output
Crossref	DOI enrichment	Bibliographic JSON
OpenAlex	Discovery and citation relationships	Linked scholarly entities in JSON
CORE	Repository content workflows	Metadata and available full text
Unpaywall	DOI-to-open-access resolution	Access-location JSON
SciSpace	Evidence comparison and research reports	Papers with metadata; cited reports, extraction tables and other files

Sources: official API documentation and SciSpace's early-access API details. (crossref.org, help.openalex.org, core.ac.uk, unpaywall.org)

Developer task map comparing four retrieval APIs with SciSpace's early-access cited research workflow

Source: editorial task map based on the documentation cited above and SciSpace's early-access API details.

How we selected these APIs

We compared documented interfaces, not independently tested uptime or recall. For your evaluation, define:

Input: DOI, title, structured filters, or a research question.
Output: records, access locations, extracted findings, or cited prose.
Coverage: known papers, missing identifiers, and inaccessible documents.
Integration: authentication, pagination, error handling, and provenance.
Acceptance: what makes a result useful to your user?

Keep metadata completeness separate from answer quality. A complete bibliographic record does not, by itself, demonstrate that an application's summary accurately represents the paper.

1. Crossref API: best for DOI metadata

What it is and coverage: Crossref exposes scholarly metadata deposited by members and trusted sources. Best for: DOI enrichment, reference matching, and bibliographic screens. GET /works/{doi} returns JSON; public access needs no key, and a valid mailto contact identifies polite-pool requests.

Pricing and limits, checked October 21, 2026: Public access is $0 per 1,000 calls. Public limits are five single-record or one list request per second; polite-pool limits are ten or three, respectively. Inspect response headers rather than applying one setting to every operation.

Boundary: Crossref supplies deposited metadata, not a paper-reading workflow. Field availability depends on deposits.

Task	Crossref	SciSpace
Enrich a DOI	Direct metadata lookup	Not a DOI metadata service; use Crossref
Explain findings	Build reading and synthesis	Read papers and return a cited report

Source: Crossref documentation and SciSpace's early-access API details.

See our Crossref works endpoint guide.

2. OpenAlex API: best for discovery and citation relationships

What it is and coverage: OpenAlex exposes works, authors, institutions, sources, and citation relationships. Best for: graph-oriented discovery. GET /works?search=... returns JSON; an API key is recommended and can be supplied through api_key.

Pricing and limits, checked October 21, 2026: Per 1,000 calls, single-entity retrieval is free, list/filter operations are $0.10, and search is $1. Accounts receive $1 daily usage; throughput is ten requests per second without a key or thirty with a free key. Reranking and content downloads are separately charged.

Boundary: OpenAlex offers search, reranking, and content access, but your product still needs a workflow for extracting findings and composing grounded comparisons.

Task	OpenAlex	SciSpace
Explore scholarly relationships	Query entities and citation links	Citation chasing forwards and backwards from seed papers
Compare methods	Build extraction and synthesis	Extraction table with AI-filled columns and a cited report

Source: OpenAlex documentation and SciSpace's early-access API details. SciSpace does not query OpenAlex.

Continue with our OpenAlex Python tutorial.

Minimal Crossref and OpenAlex requests

Set a real contact email and your OpenAlex key. These retain the essential lookup and search calls.

3. CORE API: best for repository content workflows

What it is and coverage: CORE aggregates metadata and full text from repositories and journals. Best for: an open access research api integration where document content matters. Its official client uses GET /v3/search/works, q, limit, bearer authentication, and JSON responses.

Pricing and limits, checked October 21, 2026: Eligible free access is $0 per 1,000 calls, subject to terms. Faster access requires a quote. The service page lists one batch request or five single requests per ten seconds; confirm your account's allowance.

Boundary: Check license eligibility before commercial deployment. During evaluation, distinguish a search hit from successfully retrieved, usable text.

Task	CORE	SciSpace
Retrieve repository content	Metadata and available full text	Reads 50M+ open-access full texts plus your own PDFs
Extract study findings	Build document-processing logic	Extraction table with AI-filled columns from full text

Source: CORE documentation and SciSpace's early-access API details. SciSpace does not query CORE.

4. Unpaywall API: best for DOI-to-OA resolution

What it is and coverage: Unpaywall resolves supported Crossref DOI records to open access information. Best for: access resolution after discovery. GET /v2/{doi} returns JSON and requires an email parameter.

Pricing and limits, checked October 21, 2026: REST access is $0 per 1,000 calls, with a published ceiling of 100,000 calls daily.

Boundary: An access location is not PDF bytes. url_for_pdf can be null, and a general URL can point to a landing page.

OpenAlex and Unpaywall share underlying data, so their OA results are not independent evidence.

Task	Unpaywall	SciSpace
Resolve open access	Return DOI access locations	Not an open access resolver; use Unpaywall
Compare paper findings	Add downloading, parsing, and synthesis	One task reads, extracts and writes a cited comparison

Source: Unpaywall documentation and SciSpace's early-access API details.

Minimal CORE and Unpaywall requests

Set CORE_API_KEY; reuse your real contact email for Unpaywall.

5. SciSpace API: best for cited research workflows

What it is: An agentic research API in two parts. The Search API takes a research question in plain language and returns papers with their metadata. The Task API takes a research task, and the agent picks the tools: search, reading, extraction and writing.

Best for: Research assistants and evidence-comparison products whose users need findings rather than records.

Coverage: SciSpace's 280M+ paper index plus Google Scholar, PubMed and arXiv, in one search across those sources. Full text comes from 50M+ open-access papers plus your own PDFs; paywalled papers are not read.

What comes back: The Task API streams its plan and progress live, then returns links to every file the agent created, such as a Markdown or Word report or an extraction table, plus any supporting text answer. Follow-ups continue in the same thread.

Pricing and access, checked October 21, 2026: Early access; $5 free credits at launch. See the access details below.

Task	Metadata API approach	SciSpace
Research execution	Assemble retrieval and processing	Submit one research task
Application work	Maintain the research pipeline	Show streamed progress and the returned files

Should you combine metadata APIs with SciSpace?

Yes, if your application needs both deterministic record handling and research answers. Our recommendation is to retain identifiers and provenance locally, then send evidence-comparison tasks to the SciSpace Task API.

Proposed application architecture separating metadata storage from a SciSpace research thread with citations and structured findings

Source: editorial architecture recommendation based on the cited API contracts and SciSpace's early-access API details.

For an academic metadata api evaluation, preserve original responses alongside normalized fields. Test version handling, missing identifiers, and unavailable full text rather than merging records solely by similar titles.

Review selected claims against their cited papers. Our research RAG API guide explores this product pattern.

What about publisher and specialist APIs?

Springer academic API: Distinguish Springer API offerings for metadata, open access, and full text.
dblp API: Evaluate publication, author, and venue search for computer-science discovery.
bioRxiv API and medRxiv API: Consider server-specific date and DOI queries when preprint versions matter.

Your first-week integration checklist

Define one user task and its acceptance criteria.
Test the relevant third-party retrieval path.
Request SciSpace API access; prepare representative research tasks.
Design thread persistence, a live progress display from the stream, and follow-up handling.
Plan checks for citation support and extracted-field completeness.
Test unavailable documents and interrupted workflows once access is available.

Choose records-first architecture for bibliographic products. For research-answer products, prioritize the workflow that turns evidence into the output users actually need.

Where the SciSpace API fits next to metadata APIs

Crossref, OpenAlex, CORE and Unpaywall stop at records, links and access locations. Reading the papers, extracting findings and writing a cited comparison is a pipeline you build on top. Keep direct metadata APIs when exact records or graph queries are the product.

The SciSpace API puts the full SciSpace Agent behind one call. It searches, reads, analyses and writes, and delivers the result as ready-to-use files.

API	What you send	What you get back
Search API	Any research question in plain language	A list of papers, each with its relevant metadata, from SciSpace's 280M+ paper index plus Google Scholar, PubMed and arXiv
Task API	A research task, such as "compare papers on retrieval augmented generation versus fine-tuning in an extraction table of datasets, metrics and limitations, then write a cited report"	Links to every file the agent created (Markdown, Word, PowerPoint and more) plus any supporting text answer

What this adds on top of metadata APIs

One integration instead of several. Describe the task in plain language and the agent picks the tools: search, screening, extraction, analysis and writing. You don't stitch together a search API, a PDF parser and an LLM.
Built for research. Papers with full metadata from SciSpace's 280M+ index plus Google Scholar, PubMed and arXiv, in one search across those sources.
Extraction tables from full text. AI-filled columns from full text, which you can then filter, sort and combine.
Finished files from one run. Markdown, Word, PowerPoint, or a LaTeX report compiled to PDF with its BibTeX.
Live streaming and follow-ups. The plan and each step stream while the task runs, the answer streams as it is written, and follow-ups continue in the same thread.

When to use which

Use Crossref, OpenAlex, CORE or Unpaywall for DOI records, graph queries, repository content and open access locations.
Use the SciSpace API when the deliverable is a cited report, extraction table or other file built from papers.

What it doesn't do

It does not query these metadata APIs. It does not call Crossref, OpenAlex, CORE, Unpaywall, Semantic Scholar, Scopus, Web of Science or Lens; keep those for record-level work.
No paywalled papers. It reads your own PDFs plus 50M+ open-access full texts.
Not a web search API. For news, product docs or general web content, use a web search API.

The SciSpace API is in early access; it isn't open yet. At launch it's open to anyone: register on SciSpace and start, with no sales call. Every account gets $5 in free API credits, enough to try both APIs on real questions. The API is separate from SciSpace consumer plans; no consumer plan includes API usage.

Frequently asked questions

Last checked

Third-party documentation checked October 21, 2026. SciSpace details reflect the API's early-access status on the same date. Examples were not executed as part of this review.

Sources

注意转载授权 · 阅读原文