name: arXiv Rate Limits
description: |
  Documented rate-limit and pacing policies for the arXiv public APIs.
  Limits apply jointly across the legacy Query API, OAI-PMH, and RSS feeds.
specification: api-commons-rate-limits-0.1
provider:
  id: arxiv
  name: arXiv
  url: https://info.arxiv.org/help/api/tou.html
modified: '2026-05-29'
policies:
  - id: legacy-api-baseline
    appliesTo:
      - api: arxiv-query
      - api: arxiv-oaipmh
      - api: arxiv-rss
    enforcement:
      type: client-etiquette
      mechanism: |
        arXiv does not return granular rate-limit headers but actively
        throttles or temporarily blocks clients that breach the documented
        cadence. Honour the limits client-side.
    limits:
      - dimension: request-rate
        max: 1
        per: 3s
        scope: per-client
      - dimension: concurrency
        max: 1
        scope: per-client
        note: Single connection at a time across all legacy APIs.
    burst:
      allowed: false
      note: |
        The bulk-data documentation alternately permits "bursts of 4 requests
        per second with a 1 second sleep, per burst" for custom harvesting
        against export.arxiv.org. The conservative one-per-three-seconds
        guidance remains the safe default.
    quotas:
      - dimension: results-per-request
        max: 2000
        scope: per-query
      - dimension: results-per-query
        max: 30000
        scope: per-query
        note: Refine queries returning more than 1,000 results.
  - id: bulk-harvest
    appliesTo:
      - api: arxiv-oaipmh
    enforcement:
      type: server-coordinated
      mechanism: Use resumptionToken sequences to paginate large ListRecords / ListIdentifiers calls.
    limits:
      - dimension: harvest-window
        note: |
          Metadata refresh occurs ~10:30pm ET Sunday-Thursday; schedule
          incremental harvests after the refresh and provide `from` (omit
          `until`) to avoid duplicate downloads.
contact: help@arxiv.org