specification: API Commons Rate Limits
specificationVersion: '0.1'
provider: Semantic Scholar
providerId: semantic-scholar
created: '2026-06-12'
modified: '2026-06-12'
description: >
  The Semantic Scholar API enforces rate limits at both the unauthenticated (shared pool)
  and authenticated (per-key) levels. Exceeding any limit returns HTTP 429 Too Many Requests.
  The documentation recommends implementing exponential backoff strategies for all callers.
retryAfter:
  header: Retry-After
  description: >
    When a 429 response is returned, callers should respect any Retry-After header and
    implement exponential backoff before retrying requests.
throttled:
  statusCode: 429
  description: Too Many Requests — rate limit exceeded for the current window or per-second allowance.
limits:
  - scope: unauthenticated
    metric: requests
    limit: 100
    timeFrame: 300
    unit: seconds
    notes: >
      Shared pool across all unauthenticated users. Effective throughput will be lower
      during peak usage periods.
  - scope: authenticated-api-key
    metric: requests_per_second
    limit: 1
    timeFrame: 1
    unit: seconds
    notes: >
      Introductory dedicated rate of 1 RPS across all API endpoints (Academic Graph,
      Recommendations, Datasets). Higher limits may be granted after submitting a use-case
      review request to the Semantic Scholar team.
  - scope: datasets-api
    metric: requests
    limit: null
    timeFrame: null
    unit: null
    notes: >
      Bulk dataset downloads are gated behind API key authentication. No published per-second
      rate limit; exponential backoff is still recommended for large batch jobs.
recommendations:
  - Use bulk/batch endpoints (e.g., POST /paper/batch) instead of individual lookups when
    retrieving multiple records to reduce total request count.
  - Download full corpus snapshots via the Datasets API when request volume requirements
    exceed per-key rate limits.
  - Implement exponential backoff with jitter for all production integrations.
  - Cache responses locally to avoid redundant requests against the same paper or author IDs.