{
  "_type": "curated_open_data",
  "as_of": "2026-07",
  "links": {
    "canonical": "https://verticalmarketplace.ai",
    "docs_for_llms": "https://verticalmarketplace.ai/llms.txt",
    "sell_your_own": "https://verticalmarketplace.ai/api/marketplace/listings",
    "vertical_listings": "https://verticalmarketplace.ai/api/marketplace/listings?vertical=ai-prompts"
  },
  "records": [
    {
      "unit": "ratio",
      "value": "1 token is about 0.75 words",
      "metric": "Approximate token-to-word ratio, English",
      "source": "OpenAI tokenizer guidance",
      "category": "tokenization"
    },
    {
      "unit": "ratio",
      "value": "1 token is about 4 characters",
      "metric": "Approximate token-to-character ratio, English",
      "source": "OpenAI tokenizer guidance",
      "category": "tokenization"
    },
    {
      "unit": "parameter range",
      "value": "0 to 2",
      "metric": "Sampling temperature typical range",
      "source": "OpenAI API reference",
      "category": "sampling"
    },
    {
      "unit": "parameter range",
      "value": "0 to 1",
      "metric": "Nucleus sampling top_p range",
      "source": "OpenAI API reference",
      "category": "sampling"
    },
    {
      "unit": "subjects",
      "value": "57",
      "metric": "MMLU subject count",
      "source": "Hendrycks et al. 2021, MMLU",
      "category": "benchmark"
    },
    {
      "unit": "questions",
      "value": "15,908",
      "metric": "MMLU total questions",
      "source": "Hendrycks et al. 2021, MMLU",
      "category": "benchmark"
    },
    {
      "unit": "problems",
      "value": "164",
      "metric": "HumanEval coding problems",
      "source": "Chen et al. 2021, HumanEval",
      "category": "benchmark"
    },
    {
      "unit": "problems",
      "value": "8,500",
      "metric": "GSM8K grade-school math problems",
      "source": "Cobbe et al. 2021, GSM8K",
      "category": "benchmark"
    },
    {
      "unit": "score",
      "value": "0 to 100",
      "metric": "BLEU score range",
      "source": "Papineni et al. 2002",
      "category": "evaluation metric"
    },
    {
      "unit": "metric definition",
      "value": "recall-oriented overlap for summarization",
      "metric": "ROUGE metric focus",
      "source": "Lin 2004",
      "category": "evaluation metric"
    },
    {
      "unit": "role set",
      "value": "system, user, assistant",
      "metric": "Standard chat message roles",
      "source": "OpenAI Chat API",
      "category": "prompting"
    },
    {
      "unit": "benchmark definition",
      "value": "commonsense sentence completion",
      "metric": "HellaSwag task focus",
      "source": "Zellers et al. 2019",
      "category": "benchmark"
    }
  ],
  "sources": [
    {
      "url": "https://arxiv.org/abs/2009.03300",
      "name": "MMLU paper (arXiv:2009.03300)"
    },
    {
      "url": "https://arxiv.org/abs/2107.03374",
      "name": "HumanEval paper (arXiv:2107.03374)"
    },
    {
      "url": "https://arxiv.org/abs/2110.14168",
      "name": "GSM8K paper (arXiv:2110.14168)"
    }
  ],
  "category": "benchmarks",
  "vertical": "ai-prompts",
  "data_note": "All records are public-domain facts compiled from the cited sources as of the asOf date. This content is authored and served by the platform itself — it is not seller data, so the marketplace's zero-storage promise about seller datasets is unaffected.",
  "record_count": 12,
  "what_this_is": "A platform-published open-data listing curated by Open Data Desk, the marketplace's in-house public-data seller. It is real free inventory: it counts in marketplace statistics and is purchasable for $0 through the normal purchase flow, which delivers this payload with an Ed25519-signed receipt.",
  "record_schema": {
    "unit": "Unit or basis",
    "value": "Published value or range",
    "metric": "Name of the reference figure",
    "source": "Provider docs or research paper",
    "category": "Domain the figure belongs to"
  },
  "buyer_use_cases": [
    "Estimate token budgets and cost for prompt designs",
    "Interpret published model scores on standard benchmarks",
    "Set sensible sampling parameters for a generation task"
  ]
}