Best LLMs for Profiled Document Structure Extraction

Output schema

Every answer on this task is checked against this JSON Schema, whichever model wrote it. An answer that doesn't fit counts as a model failure, and the call is retried on another model.

{
  "$defs": {
    "TOCChapter": {
      "description": "A chapter identified in the table of contents.",
      "properties": {
        "page_number": {
          "anyOf": [
            {
              "type": "integer"
            },
            {
              "type": "null"
            }
          ],
          "description": "Page number where the chapter starts, if available",
          "title": "Page Number"
        },
        "section_number": {
          "anyOf": [
            {
              "type": "string"
            },
            {
              "type": "null"
            }
          ],
          "description": "Section number (e.g., 'Item 1', 'Part I')",
          "title": "Section Number"
        },
        "title": {
          "description": "The full title of the chapter/section",
          "title": "Title",
          "type": "string"
        }
      },
      "required": [
        "title",
        "page_number",
        "section_number"
      ],
      "title": "TOCChapter",
      "type": "object"
    }
  },
  "description": "Table of contents extracted from S-1 filing.",
  "properties": {
    "chapters": {
      "description": "List of chapters/sections found in the table of contents",
      "items": {
        "$ref": "#/$defs/TOCChapter"
      },
      "title": "Chapters",
      "type": "array"
    },
    "document_type": {
      "description": "Type of document (S-1, S-1/A, etc.)",
      "title": "Document Type",
      "type": "string"
    },
    "toc_found": {
      "description": "Whether a clear table of contents was found",
      "title": "Toc Found",
      "type": "boolean"
    }
  },
  "required": [
    "chapters",
    "document_type",
    "toc_found"
  ],
  "title": "TableOfContents",
  "type": "object"
}