Evidence · first-party tested

The parser recovers the opening content but loses the distinction between the document title and the abstract, reducing the semantic organization of the parsed page.

⚠ Struggled🧾 artifact-verifiedTested Jun 20, 2026Mistral AI →
Output — unretouched
Mistral AI — scanned_pdf_page_1.png
Mistral AI — scanned_pdf_page_1.png
Output — unretouched
Parsed text hierarchy from the research note title page.
Parsed text hierarchy from the research note title page.
The observation

The parser recovers the opening content but loses the distinction between the document title and the abstract, reducing the semantic organization of the parsed page.

Criterion: Reading Order & Structure · Scenario: Scanned Research Paper (group: scanned-research-paper)
Query this
get_evidence({
  tool: "mistral-ai",
  scenario: "scanned-research-paper"
})
MCP · mcp.aidemos.com/api/mcp
Free with attribution.
Same input, other tools
Adobe PDF Extract API ✗ FailedExtend AI ✓ WorkedLanding AI ✗ FailedLlamaparse ✓ WorkedNutrient.io ✗ FailedTensorlake ✓ WorkedUpstage AI ✗ Failed
Real outputs, no retouching · every cell queryable via API & MCP · aidemos.com