specification: FinOps Framework
specificationVersion: '1.0'
provider: Semantic Scholar
providerId: semantic-scholar
created: '2026-06-12'
modified: '2026-06-12'
description: >
  Semantic Scholar provides its Academic Graph API, Recommendations API, and Datasets API
  at no direct monetary cost to API consumers. The FinOps considerations are therefore
  centered on indirect costs: engineering time, infrastructure for local caching or dataset
  storage, compute for processing bulk corpus downloads, and organizational risk management
  related to rate-limit-induced latency or key revocation.
billingModel: free
focusColumns:
  - ChargeType: Usage
  - ChargeCategory: None
  - BilledCurrency: N/A
  - ConsumedUnit: API Requests
  - ConsumedQuantity: variable
  - ListUnitPrice: 0.00
  - EffectiveUnitPrice: 0.00
  - InvoiceIssuerName: Allen Institute for AI (AI2)
  - ServiceName: Semantic Scholar Academic Graph API
  - ServiceCategory: Analytics
  - ResourceType: REST API
meters:
  - name: API Requests (Unauthenticated)
    unit: requests
    price: 0.00
    limit: 100 per 5 minutes (shared pool)
    notes: No cost; indirect cost is shared contention and unpredictable latency.
  - name: API Requests (Authenticated / API Key)
    unit: requests_per_second
    price: 0.00
    limit: 1 RPS dedicated
    notes: Free API key; indirect costs include key management and potential review delays.
  - name: Dataset Download
    unit: corpus_snapshot
    price: 0.00
    limit: monthly snapshots with diff support
    notes: >
      Bulk corpus files may be large (multi-GB). Indirect costs include egress bandwidth,
      storage, and compute for data processing pipelines.
principles:
  - name: Visibility
    description: >
      Track API usage volume and error rates (especially 429s) using application-level
      metrics. Monitor the status page at status.api.semanticscholar.org to anticipate
      service disruptions.
    actions:
      - Instrument all API calls with logging for response codes, latency, and retry counts.
      - Set up alerting on elevated 429 rate to detect approaching rate limit boundaries.
      - Review API release notes and the s2-folks GitHub repository for breaking changes.
  - name: Optimization
    description: >
      Minimize requests through caching, batching, and selective field retrieval to stay
      within rate limits and reduce downstream compute costs.
    actions:
      - Use POST /paper/batch and equivalent bulk endpoints instead of sequential GETs.
      - Request only required fields using the `fields` query parameter to reduce payload size.
      - Cache paper and author records locally; re-query only when freshness is required.
      - For high-volume workloads, use the Datasets API bulk corpus instead of per-record API calls.
      - Implement exponential backoff to avoid wasted retry requests during throttling windows.
  - name: Risk Management
    description: >
      API key availability is not guaranteed — keys are pruned after ~60 days of inactivity
      and new key approval is subject to review backlogs. Plan for continuity risks.
    actions:
      - Keep API keys active with at least one request every 30 days.
      - Maintain a local dataset snapshot as a fallback for critical workloads.
      - Do not depend on third-party application key access; each production system should
        hold its own institutional-email-registered key.