arXiv API 使用指南：Python 调用、API Key、定价与速率限制

The arXiv API is a free interface for searching paper metadata and abstracts; its documented quickstart needs no API key and returns Atom XML, according to API Basics and the terms. With requests installed, this Python example queries the official search endpoint and prints one normalized title.

Output:

Tested Oct 22, 2026.

Run examples one at a time, wait at least three seconds between runs, and do not run other arXiv clients concurrently, following the shared request limit in the terms. These shortened examples stop on HTTP errors rather than retrying automatically.

arXiv API at a glance

API key, pricing and rate limits

API key and pricing

No key is used by the documented public-search quickstart in API Basics. This applies to the search interface, not every arXiv service.

The official terms describe “Providing free Application Programming Interfaces (APIs)”. Bulk full-text transfer is separate: the S3 guide states that “the downloader pays Amazon for the download based on bandwidth used”.

Request spacing and refreshes

The terms specify “one request every three seconds, and limit requests to a single connection at a time.” The restriction applies collectively across machines you control. Share a request queue rather than creating a limiter per worker.

The User's Manual describes a “24 hour submission cycle”. Cache repeated production queries and refresh around the daily cycle rather than repeatedly polling an unchanged search.

Search queries and pagination

The arXiv API combines fielded queries with zero-based start offsets and max_results page sizes, as documented in the User's Manual. Pass parameters through requests rather than manually assembling an encoded URL.

Read the total for one page

This self-contained example reads opensearch:totalResults, checks for Atom error entries and prints a compact page summary using the response fields in the User's Manual.

Output:

Conflicting result-cap documentation: the official User's Manual says “30000 in slices of at most 2000”, while the arxiv.py documentation says “300,000 results per query”. Plan around the official ceiling. For bulk ingestion, use the recommended metadata interface.

Python libraries

The arxiv package is a maintainer-published wrapper; feedparser is an optional feed parser rather than a dedicated arXiv SDK, as described on arxiv's PyPI page and feedparser's PyPI page.

Migrate older arxiv.py examples

Set Search.max_results explicitly. For submission-date ordering, the package example on PyPI uses arxiv.SortCriterion.SubmittedDate.

Metadata and bulk access

Convert search records to JSONL

The search response is Atom XML, so JSON is an application-side conversion, as established in API Basics. This self-contained example normalizes titles and abstracts, preserves the returned identifier, and stores the PDF URL without downloading the document.

Output:

Keep these distinctions in stored records:

• Versioned identifiers: an unversioned identifier refers to the latest version. Preserve the returned version to identify the exact paper used. See the identifier guide.
• DOI metadata: the author-supplied DOI identifies another version, such as a journal publication, rather than the arXiv-assigned DOI. It can contain space-separated DOIs. See the metadata guidance.
• PDF references: a URL is not a reuse license. Check the work's license before storing or redistributing files. See the reuse guidance.

Move bulk synchronization to OAI-PMH

arXiv recommends OAI-PMH for bulk metadata downloads and maintaining an updated local copy in its bulk-data guide. The March 2025 rewrite changed its base URL to https://oaipmh.arxiv.org/oai, according to the migration notes.

OAI-PMH datestamps track metadata changes, including administrative updates; the interface does not support selective harvesting by submission date, according to the OAI-PMH guide. For original submission dates, use search's submittedDate filter from the User's Manual.

Common errors and terms

Successfully parsing XML does not establish that every entry is a paper: arXiv can return an error as an Atom entry, as explained in the User's Manual.

Terms and compliance

Use machine-access interfaces rather than indiscriminate crawling: the robots guidance states that “Indiscriminate automated downloads from this site are not permitted”. A managed scraper does not remove that obligation.

Metadata and article-file rights are separate. The default distribution license does not directly grant downstream reuse permissions, according to the reuse guidance. Check the paper's license or obtain permission before reposting full text.

The API Access page requests this acknowledgment: “Thank you to arXiv for use of its open access interoperability.” Do not imply endorsement through promotional use of arXiv's name or marks without written authorization; see the brand guidelines.

arXiv API alternatives

Choose an alternative for a different source or managed execution, not to bypass arXiv's usage rules.

From arXiv metadata to finished research: the SciSpace API

The arXiv API tells you which papers exist. It doesn't read them, compare them or write anything up. To turn search results into a literature review, you still have to fetch permitted full text, parse PDFs, extract findings, format citations and add sources beyond arXiv yourself.

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.

It comes as two APIs:

What this adds on top of the arXiv API

• One search across arXiv and beyond. The Search API covers arXiv along with SciSpace's index, Google Scholar and PubMed, so you don't maintain a client for each source. Results come back as paper records with full metadata, not Atom XML.
• A finished write-up, not a list of links. A single Task API run can return a LaTeX report compiled to PDF with its BibTeX, a Word document, a PowerPoint deck, plots or a poster. You get links to every file it created.
• Progress you can show users. Research runs can take minutes. The Task API streams each plan step and the answer as it is written, so your app can show live progress instead of a spinner. When it finishes, send follow-ups in the same thread to refine, extend or reformat the work.
• Questions over specific papers. Upload up to 10 PDFs at once and get answers with page-level citations, including content read from figures and tables.

What it doesn't do

• No paywalled papers. SciSpace reads your own PDFs plus 50M+ open-access full texts. It doesn't bypass publisher access.
• Not a bulk-metadata mirror. To keep a local copy of all arXiv metadata, use OAI-PMH as described above.
  

Access

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

Does arXiv have an official API?

Yes. Start with arXiv's official API Access page, rather than assuming similarly named third-party services are operated by arXiv.

Does the arXiv API return JSON?

No. It returns Atom XML, according to API Basics. Parse the feed, then serialize the fields you need.

How do I use the arXiv API with Python?

Fetch and parse the feed with requests, or construct arxiv.Client and arxiv.Search and iterate client.results(search), following the package documentation.

What should I use to download all arXiv metadata?

Use OAI-PMH for bulk downloads and synchronization, as recommended in the bulk-data guide.

Last checked

Facts, prices and code examples in this guide were checked on October 22, 2026.

Sources

1. arXiv API Basics
2. arXiv API User's Manual
3. arXiv API Terms of Use
4. arXiv Bulk Data Access
5. arXiv Bulk Data Access via Amazon S3
6. arxiv Python package
7. feedparser Python package
8. arxiv.py API documentation
9. arxiv.py 4.0.0 release notes
10. arXiv Identifier Guide
11. arXiv Submission Preparation and Metadata
12. arXiv OAI-PMH Interface
13. arXiv Robots Guidance
14. arXiv Reuse Permissions
15. arXiv API Access
16. arXiv Brand Guidelines
17. Perconey arXiv Scraper on Apify
18. Perconey arXiv Scraper Pricing
19. OpenAlex API Documentation
20. OpenAlex Access and Pricing
21. Crossref REST API Documentation

注意转载授权 · 阅读原文