Semantic Scholar references API

Use the Semantic Scholar references API to retrieve referenced papers and the citation relationships published by the source. Provide the citing paper’s identifier and optionally filter intent or ordering to inspect the papers it references, their citation contexts and influence flags in the backward citation direction. Trace the sources a paper relies on, inspect its citation context, or build a backward-reading list from a known publication. References follow the selected record’s backward citation relationships; the amount of bibliographic detail depends on the source endpoint. The detail page provides request and response examples, supported parameters, and usage pricing.

Selected price
$0.002 / billing creditPublished pricing
Sellers / implementations
1 / 1
Requests · 30d
—Loading...
Success rate · 30d
—Loading...

Run sends a real API request. Charges follow the published API contract.

Request parameters
Paper id: a Semantic Scholar id (40 hex characters), `CorpusId:<n>`, `DOI:<doi>`, `ARXIV:<id>`, `PMID:<n>`, `PMCID:<id>`, `MAG:<n>`, `ACL:<id>`, `URL:<link>`, a bare DOI or arXiv id, or a semanticscholar.org / doi.org / arxiv.org link.
Optional parameters (8)
Order.
Only references with this intent.
Page number, from 1.
0 leaves out the bulky `raw` source blocks (often 10-30x smaller).
Comma-separated fields to return, e.g. `posts.id,posts.title`. Call metadata is always included.
1 skips the 24-hour cache and fetches live.
JSON only for this OpenAny product. Omit output or use json; CSV and NDJSON exports are not exposed.
Native CSV/NDJSON record selector; it has no effect on this JSON-only product.

Sample response

{
  "attribution": "Data: Semantic Scholar (Allen Institute for AI)",
  "cached": false,
  "count": 10,
  "credits_used": 1,
  "id": "ARXIV:1706.03762",
  "intent": "all",
  "intent_counts": {
    "background": 19,
    "methodology": 13,
    "result": 2
  },
  "items": [
    {
      "abstract": "Several mechanisms to focus attention of a neural network on selected parts of its input or memory have been used successfully in deep learn…",
      "authors": [
        {
          "author_id": "40527594",
          "name": "Lukasz Kaiser"
        },
        {
          "author_id": "1751569",
          "name": "Samy Bengio"
        }
      ],
      "citation_count": 62,
      "contexts": [
        {
          "intents": [],
          "text": "We trained our models on one machine with 8 NVIDIA P100 GPUs."
        },
        {
          "intents": [],
          "text": "We estimate the number of floating point operations used to train a model by multiplying the training time, the number of GPUs used, and an …"
        },
        "… 3 more"
      ],
      "corpus_id": 11250862,
      "fields_of_study": [
        "Computer Science"
      ],
      "is_influential": true,
      "paper_id": "735d547fc75e0772d2a78c46a1cc5fad7da1474c",
      "reference_count": 27,
      "title": "Can Active Memory Replace Attention?",
      "tldr": "An extended model of active memory is proposed that matches existing attention models on neural machine translation and generalizes better t…",
      "url": "https://www.semanticscholar.org/paper/Can-Active-Memory-Replace-Attention-Kaiser-Bengio/735d547fc75e0772d2a78c46a1cc5fad7da1474c",
      "venue": "NIPS",
      "year": 2016
    },
    {
      "abstract": "We introduce a neural network with a recurrent attention model over a possibly large external memory. The architecture is a form of Memory N…",
      "authors": [
        {
          "author_id": "2265067",
          "name": "Sainbayar Sukhbaatar"
        },
        {
          "author_id": "3149531",
          "name": "A. Szlam"
        },
        "… 2 more"
      ],
      "citation_count": 2807,
      "contexts": [
        {
          "intents": [
            "background"
          ],
          "text": "End-to-end memory networks are based on a recurrent attention mechanism instead of sequencealigned recurrence and have been shown to perform…"
        }
      ],
      "corpus_id": 1399322,
      "fields_of_study": [
        "Computer Science"
      ],
      "is_influential": false,
      "paper_id": "4f10b9f47c5bb6b54dd4f5ca8d9fa2c0bbd7ec5e",
      "reference_count": 25,
      "title": "End-To-End Memory Networks",
      "tldr": "A neural network with a recurrent attention model over a possibly large external memory that is trained end-to-end, and hence requires signi…",
      "url": "https://www.semanticscholar.org/paper/End-To-End-Memory-Networks-Sukhbaatar-Szlam/4f10b9f47c5bb6b54dd4f5ca8d9fa2c0bbd7ec5e",
      "venue": "NIPS",
      "year": 2015
    },
    "… 8 more"
  ],
  "next_page": 2,
  "page": 1,
  "paper_id": "204e3073870fae3d05bcbc2f6a8e263d9b72e776",
  "sort": "relevance",
  "total": 41,
  "total_pages": 5
}

Response result

Response structure

Inferred from examples
  • attributionstring
  • cachedboolean
  • countinteger
  • credits_usedinteger
  • idstring
  • intentstring
  • intent_countsobject
    • backgroundinteger
    • methodologyinteger
    • resultinteger
  • items(object | string)[]
    • abstractstring
    • authors(object | string)[]
      • author_idstring
      • namestring
    • citation_countinteger
    • contexts(object | string)[]
      • intentsstring[]
      • textstring
    • corpus_idinteger
    • fields_of_studystring[]
    • is_influentialboolean
    • paper_idstring
    • reference_countinteger
    • titlestring
    • tldrstring
    • urlstring
    • venuestring
    • yearinteger
  • next_pageinteger
  • pageinteger
  • paper_idstring
  • sortstring
  • totalinteger
  • total_pagesinteger

Common questions

Do I need a separate key for every provider?

Use your OpenAny API key for standard API access. Read it from OPENANY_API_KEY in your environment and keep it out of prompts, source code and logs.

Are these providers interchangeable?

No. Each offering has its own inputs, pricing and response contract. Selecting a catalog offering does not establish automatic routing or fallback. Confirm the service contract before calling.

How is this API billed?

API-key calls use your shared balance at the selected implementation's published rate. Check the billing unit and request-specific pricing rules before making a paid call. Recorded usage determines the final charge under that contract.

Where do the price and contract come from?

The selected implementation publishes its provider, request contract, revision and billing rules through OpenAny's public API catalog. A revision number identifies a contract version, not a price update date. A price update timestamp is not supplied; do not treat a page date as one.

How do OpenAny Balance and x402 differ?

OpenAny Balance uses an API key and one shared balance. Card, USDC and x402 top-ups, card auto top-up and USDC subscriptions fund that balance; available routes appear on the Credits page. Direct x402 is a separate pay-per-request flow, available only on listings that show it and requiring an authorized payment client.

Does Playground send a paid request?

Yes. After you sign in, Run sends a real API request using an active OpenAny API key. Charges follow the published API contract and use your shared balance. Selecting an implementation, editing inputs and viewing sample responses do not send requests.

Use Semantic Scholar references API with your agent

Inspect endpoint contracts and confirm availability and pricing before a paid call.