developer-tools · tested 2026-06-20

Best AI APIs to Convert Complex PDFs to Clean Markdown

We tested hosted PDF-to-markdown APIs on the same three hard documents: a long hybrid annual report, a table-heavy financial report, and an image-only scanned research paper. The goal was usable markdown with OCR, tables, charts, and reading order preserved well enough for downstream RAG, search, and reuse.

0
8 tools7 things we checked3 tests809 findings1514 screenshots1 recordings67 output files14 min read

The ranking

Scores are the average across every check we scored for that tool. Not every tool was scored on every check — the count is shown.

ToolScoreWhere it lands
#1Extend AIBest3.8/5
7 checks
Strong hybrid-document parsing, but visuals often stay out of flow
#2LlamaParseBest3.3/5
7 checks
Strong on reading order and OCR for mixed PDFs, but weaker on visual retention and complex table semantics.
#3Landing AIBest3.0/5
7 checks
Strong at table-heavy document reconstruction, but weaker on visual fidelity and heading semantics.
#4Mistral AIUsable3.4/5
7 checks
Strong OCR and export automation, with good table recovery but inconsistent hierarchy on longer documents.
#5TensorlakeUsable3.6/5
7 checks
Strong document structure and table parser, but weak on hierarchical scanned tables
#6Adobe APIUsable3.4/5
7 checks
Best at keeping visual assets and financial tables in place; weaker on signatures and hierarchy.
#7Upstage AINeeds work2.7/5
7 checks
Strong at native financial table reconstruction, but weak on scanned multicolumn structure and visual preservation.
#8Nutrient.ioNeeds work2.0/5
7 checks
Good at basic OCR and section hierarchy, but weak on tables, charts, and other visual content in complex documents.

What we checked

Every finding below is tied to one of these checks, and to the test that produced it. The number is how many of the 8 tools we recorded findings for.

Complex Document Handling 8 toolsReading Order & Structure 8 toolsTable Preservation 8 toolsVisual Content Retention 8 toolsText & OCR Completeness 6 toolsAdvanced Features (Bonus) 5 toolsMarkdown Quality 5 tools

What we tried

The same 3 tests were run on every tool.

Scanned Research PaperSumitomo Heavy Industries Consolidated Financial ReportTarget 2015 Annual Report
Read it

Extend AI

Best#1 of 8

Strong hybrid-document parsing, but visuals often stay out of flow

Complex Document Handling4/53 findings

Handles long mixed-content PDFs well, though quality drops somewhat on the most complex multilevel table layouts.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Handles an 84-page mixed-content financial report end to end in a single automated markdown export, keeping the converted document usable as a coherent report rather than degrading across sections.

llamaparse-hybrid-earnings-pdf-1.pdf
LlamaParse — Hybrid-Earnings-PDF.pdf
extend-ai-extendai-hybrid-earnings-pdf-output-6.md
Loading file...
Extend AI — extendai_hybrid_earnings_pdf_output.md
Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Maintains usable structure across an 84-page mixed-content document that combines narrative text, tables, charts, and scanned marks without obvious degradation.

llamaparse-hybrid-earnings-pdf-1.pdf
LlamaParse — Hybrid-Earnings-PDF.pdf
extend-ai-extendai-hybrid-earnings-pdf-output-6.md
Loading file...
Extend AI — extendai_hybrid_earnings_pdf_output.md
Reading Order & Structure4.5/517 findings

Keeps section hierarchy and reading flow clear across long reports and scanned papers, with only limited structural drift around complex tables.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Keeps section hierarchy and explanatory flow intact in an 84-page hybrid annual report, so the markdown remains readable as a report rather than a flat text dump.

llamaparse-hybrid-earnings-pdf-1.pdf
LlamaParse — Hybrid-Earnings-PDF.pdf
extend-ai-extendai-hybrid-earnings-pdf-output-6.md
Loading file...
Extend AI — extendai_hybrid_earnings_pdf_output.md
Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Maintains report hierarchy and reading flow across an 84-page hybrid annual report, keeping headings, narrative paragraphs, and bullets in a coherent markdown structure instead of flattening the document.

extend-ai-extendai-hybrid-earnings-pdf-output-6.md
Loading file...
Extend AI — extendai_hybrid_earnings_pdf_output.md
Table Preservation3.5/527 findings

Preserves row/column alignment and grouped headers well overall, but multirow and compound headers break in some cases.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Preserves grouped-column hierarchy in multilevel tables, carrying header and subheader relationships into the extracted markdown.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Reconstructs tabular content with aligned rows and columns, preserving the original financial table structure in markdown.

Visual Content Retention3/518 findings

Charts and logos are retained as captions or references, but they are not consistently kept inline with the document flow.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Retains chart meaning by converting a waterfall graphic into a captioned structured element that explains the visual progression in text.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Detects lightly visible handwritten markings and carries them into the output, showing retention of subtle low-contrast content from the scan.

Text & OCR Completeness4.5/511 findings

Covers native text, scanned pages, signatures, handwriting, and low-clarity stamps with only minor OCR slips and a few missing contextual bits.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Captures lightly visible handwritten or marked content and carries it into the extracted representation.

Mixedwhen we tried: Target 2015 Annual Reportlink to this finding

Extracts low-clarity signature and stamp text, but can introduce small OCR errors such as reading 'LLP' as '1LP'.

Advanced Features (Bonus)2/514 findings

Shows structured table/chart extraction, but there is no clear evidence of explicit low-confidence OCR or ambiguity flagging.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Provides separate chart extraction by turning chart content into a captioned figure element that preserves the chart's meaning beyond the visual encoding.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Adds a caption element for chart extraction, turning the chart into structured text while keeping the numeric relationships legible.

LlamaParse

Best#2 of 8

Strong on reading order and OCR for mixed PDFs, but weaker on visual retention and complex table semantics.

Complex Document Handling4/55 findings

Handled long hybrid and table-heavy documents consistently, with only limited degradation in the more complex table regions.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Handles an 84-page hybrid annual report end-to-end and returns a usable markdown output without manual correction or post-processing.

llamaparse-hybrid-earnings-pdf-1.pdf
LlamaParse — Hybrid-Earnings-PDF.pdf
llamaparse-llamaparse-target-earnings-output-1.md
Loading file...
LlamaParse — llamaparse_target_earnings_output.md
Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Handles an 84-page mixed-content annual report end-to-end and returns a usable markdown export without manual correction or post-processing.

llamaparse-hybrid-earnings-pdf-1.pdf
LlamaParse — Hybrid-Earnings-PDF.pdf
llamaparse-llamaparse-target-earnings-output-1.md
Loading file...
LlamaParse — llamaparse_target_earnings_output.md
Reading Order & Structure4/530 findings

Kept headings, section flow, and multi-column reading order recognizable across the test documents, with some loss in structured sublayouts.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Reconstructs a scanned multi-column paper into a coherent reading flow, keeping section headings attached to the paragraphs that follow.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Preserves section hierarchy and overall content flow in a multi-column annual report instead of breaking the page into disconnected reading fragments.

Table Preservation3/549 findings

Preserved standard tables and much of the visible data, but multi-level headers, grouped relationships, and TOC structure were only partially retained.

Struggledwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

For a complex grouped-header table, the output keeps the visible values but weakens parent-child column relationships, making the structure less explicit.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Retains a standard financial table’s row alignment, column organization, and value associations in the markdown output.

Visual Content Retention2/519 findings

Did not truly retain visuals inline; charts and assets were mostly converted into text or tables rather than preserved as images in place.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Converts a chart into a structured table while preserving legend-to-value mapping, so the underlying data relationships remain intact.

Struggledwhen we tried: Scanned Research Paperlink to this finding

Does not retain charts as charts; the chart is converted into a structured table even though legend-to-value mapping survives.

Text & OCR Completeness4/51 finding

Recovered the readable content well across scanned and hybrid PDFs, with only some structure-related losses in complex areas.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Extracts embedded signature text from an image-based signature block, recovering the readable 'Ernst & Young LLP' text from the visual asset.

Advanced Features (Bonus)2/514 findings

Showed some extra handling for charts and assets, but there was no clear evidence of low-confidence OCR or ambiguous-region flagging.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Provides separate chart extraction by converting a 9-step SG&A waterfall chart into a compact parsed table that retains the three SG&A rate points (20.2%, 20.0%, 19.6%) and the intermediate component deltas.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Provides separate chart extraction by turning a mortality bar chart into a structured table with 3 years of values across 5 categories, while retaining the legend-to-value mapping.

Markdown Quality4/52 findings

Produced usable downloadable markdown rather than a flat dump, though some extracted structures were simplified or flattened.

Struggledwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Extracts the table of contents as sequential text rather than a structured TOC, so entries and page numbers are recovered but the TOC’s organization is lost.

Struggledwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Extracts a table of contents as sequential text rather than a structured markdown block, so the entries and page numbers are recovered but the layout hierarchy is lost.

Landing AI

Best#3 of 8

Strong at table-heavy document reconstruction, but weaker on visual fidelity and heading semantics.

Complex Document Handling4/54 findings

Handled long, mixed-content financial and scanned documents well overall, with some hierarchy degradation on harder sections.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Processes an 84-page hybrid report with tables, charts, and scanned signatures in one automated pass, producing usable markdown without manual correction.

llamaparse-hybrid-earnings-pdf-1.pdf
LlamaParse — Hybrid-Earnings-PDF.pdf
landing-ai-landingai-hybrid-earningspdf-output.md
Loading file...
Landing AI — landingai_hybrid_earningspdf_output.md
Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Handles an 84-page hybrid financial report containing tables, charts, and scanned signatures, returning usable markdown without manual correction.

llamaparse-hybrid-earnings-pdf-1.pdf
LlamaParse — Hybrid-Earnings-PDF.pdf
landing-ai-landingai-hybrid-earningspdf-output.md
Loading file...
Landing AI — landingai_hybrid_earningspdf_output.md
Reading Order & Structure3/543 findings

Section flow and local hierarchy often held up, but top-level headings and opening-page structure were inconsistently preserved.

Failedwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Flattens the document's highest-level heading into plain text instead of an H1, which weakens the reconstructed hierarchy even though the content is retained.

Failedwhen we tried: Target 2015 Annual Reportlink to this finding

The highest-level heading is flattened into plain text instead of being preserved as an H1, which weakens the reconstructed document hierarchy.

Table Preservation4/533 findings

Rebuilt tables well with rows, columns, and headers mostly intact, though nested header distinctions were sometimes collapsed.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Reconstructs a financial table in markdown with its header row, row labels, and numeric columns aligned to the source layout.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Successfully reconstructs a multilevel financial table in markdown, preserving header relationships, row alignment, and column organization.

Visual Content Retention1/516 findings

Charts, signatures, and stamps were not retained as visual assets; they were mostly converted into textual or semantic descriptions.

Mixedwhen we tried: Scanned Research Paperlink to this finding

Replaces a chart with descriptive text that carries numeric values and legend details, but does not retain the chart as a visual figure.

Failedwhen we tried: Target 2015 Annual Reportlink to this finding

Does not retain the waterfall chart as a visual asset; it converts the chart into a text description of values and increase/decrease relationships instead.

Text & OCR Completeness4/51 finding

Captured essentially all readable text, including scanned pages, but some structure and heading semantics were flattened.

Struggledwhen we tried: Scanned Research Paperlink to this finding

Fragments the vertically oriented 'cut completed' note in the table into five OCR pieces ('ed', 'et', 'np', 'cut con', and 'cut'), and the final check-area line is truncated in the extracted text.

Advanced Features (Bonus)1/519 findings

No clear separate table/chart extraction or explicit low-confidence OCR flagging, despite some semantic descriptions of signatures and stamps.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Represents the blurred "Ernst & Young" marker with a generated description, preserving an ambiguous low-visibility region semantically instead of omitting it.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

The signature region is captured as a semantic attestation element, preserving the presence and characteristics of the handwritten signature instead of only extracting the signer name.

Markdown Quality4/52 findings

Returned usable markdown with headings and tables rather than a flat dump, though some structure was simplified.

Worked wellacross all testslink to this finding

Returns parsed markdown as a downloadable output through a fully automated API workflow, with no manual correction or post-processing required.

landing-ai-landingai-hybrid-earningspdf-output.md
Loading file...
Landing AI — landingai_hybrid_earningspdf_output.md
landing-ai-landingai-financialpdf-output.md
Loading file...
Landing AI — landingai_financialpdf_output.md
landing-ai-landingai-scannedpdf-output.md
Loading file...
Landing AI — landingai_scannedpdf_output.md
Worked wellacross all testslink to this finding

Returns the extraction as downloadable Markdown output across runs, rather than only as an in-app preview.

Mistral AI

Usable#4 of 8

Strong OCR and export automation, with good table recovery but inconsistent hierarchy on longer documents.

How it scored

Complex Document Handling3/5Reading Order & Structure3/5Table Preservation3/5Visual Content Retention4/5
Text & OCR Completeness · no findings4/5
Advanced Features (Bonus) · no findings3/5
Markdown Quality4/5
Complex Document Handling3/56 findings

Processed long mixed-content PDFs end-to-end, but quality degraded on hierarchy and complex table reconstruction in larger documents.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Handles an 84-page mixed-content annual report end-to-end in a fully automated pass, including tables, charts, and scanned signatures, without manual correction or post-processing.

mistral-ai-mistral-ai-hybrid-earnings-pdf-output-zip-3.zip

ZIP
Mistral AI — Mistral AI Hybrid Earnings PDF Output ZIP.zip
Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Supports dual output organization for a long report by creating separate page-level files alongside a consolidated markdown document, which scales localized validation across the document.

mistral-ai-mistral-ai-hybrid-earnings-pdf-output-zip-3.zip

ZIP
Mistral AI — Mistral AI Hybrid Earnings PDF Output ZIP.zip
Reading Order & Structure3/540 findings

Reading flow was often preserved, but hierarchy was inconsistent, with flattened TOCs and missed section levels.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

The scanned paper is reconstructed with section hierarchy and reading flow intact, so headings and supporting paragraphs stay correctly connected despite the multi-column layout.

Failedwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

The table of contents entries remain present, but the navigational hierarchy is flattened into raw text, so the original TOC organization is not preserved.

Table Preservation3/540 findings

Handled some layered financial tables well, but multilevel headers and complex scanned tables lost structural fidelity.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Reconstructs a 5-year financial results table into a usable markdown table, keeping year columns and row-to-value relationships intact.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Preserves layered financial tables in a usable structured form, keeping header, row, and value relationships aligned rather than collapsing the table into plain text.

Visual Content Retention4/515 findings

Charts, signatures, and other visuals were retained as page-linked assets in the output folders rather than being dropped.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

The parser exposes charts through page-wise markdown files and visual assets, keeping the extracted visual content linked to its original document location.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

The parser extracts visual assets into page-specific output folders so charts, signatures, and other elements remain associated with their source pages instead of being collapsed into one output block.

Markdown Quality4/55 findings

Exported usable overall and page-wise Markdown files in downloadable ZIPs, though structure could flatten in places.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Produces clean, usable markdown exports in both consolidated and page-level form within a single ZIP package, supporting both whole-document reading and local inspection.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Returns usable markdown as a downloadable export, including both page-level files and a consolidated document packaged together for downstream inspection.

mistral-ai-mistral-ai-financial-pdf-output-zip-file-2.zip

ZIP
Mistral AI — Mistral AI Financial PDF Output ZIP File.zip

Tensorlake

Usable#5 of 8

Strong document structure and table parser, but weak on hierarchical scanned tables

Complex Document Handling4/56 findings

Holds up across long mixed-content reports, but quality drops on the most complex scanned tables.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Handles an 18-page table-heavy financial report without breaking section ordering or structural flow, even with numerous tables distributed across the document.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Handles an 84-page mixed-content report while keeping heading order and section relationships aligned across tables, charts, and scanned signatures.

llamaparse-hybrid-earnings-pdf-1.pdf
LlamaParse — Hybrid-Earnings-PDF.pdf
Reading Order & Structure4/521 findings

Preserves section order and document hierarchy well across long hybrid and scanned documents.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Preserves document hierarchy and heading order so the extracted flow stays close to the source layout.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Preserves heading order and section relationships in an 84-page financial report, keeping the extracted hierarchy closely aligned with the source layout.

Table Preservation3/529 findings

Keeps ordinary and multi-section tables mostly intact, but multi-header and hierarchical tables lose headers and relationships.

Failedwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Fails on multi-header tables by not inferring the header hierarchy and omitting at least one header label, which yields an incomplete reconstruction.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Reconstructs financial tables with row, column, and value relationships intact, keeping the table structure structurally faithful in markdown.

Visual Content Retention3/54 findings

Extracts chart data and signature content, but visuals are represented as parsed data rather than faithfully retained images in place.

Failedwhen we tried: Target 2015 Annual Reportlink to this finding

Embedded figures are not retained as visual objects in the markdown output; they are converted into text-only figure descriptions instead of being placed back into the document as images.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Recovers handwritten signature content from scanned pages, including distinct signer entries, instead of dropping the signature imagery entirely.

Text & OCR Completeness4/513 findings

Covers scanned signatures and blurry text with few omissions, but complex scanned tables still break down.

Mixedwhen we tried: Target 2015 Annual Reportlink to this finding

Recovers degraded stamp text well enough to capture the Ernst & Young reference, but makes a symbol-level OCR error by rendering the ampersand as a plus sign.

Mixedwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

The parser can read degraded stamp text, including the Ernst & Young reference, but it introduces a one-symbol OCR error by rendering an ampersand as a plus sign.

Advanced Features (Bonus)3/516 findings

Adds separate chart extraction and signature parsing, but no explicit low-confidence OCR flags are described.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Extracts chart values into a separate tabular representation without requiring a dedicated chart-processing mode, showing an extra structured extraction path.

Worked wellacross all testslink to this finding

Exposes API-key access and documentation from the home page, giving the tool a built-in API entry point alongside the UI workflow.

Markdown Quality4/53 findings

Outputs usable copyable markdown with clear structure, though it is not a downloadable export.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Produces usable markdown with clear section headings, a figure block, and bullet lists rather than collapsing the report into a flat text dump.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Outputs usable markdown rather than a flat text dump, with section headings, table blocks, chart blocks, and signature text separated into readable markdown structure.

tensorlake-tensorlake-hybrid-earningspdf-output.md
Loading file...
Tensorlake — tensorlake_hybrid_earningspdf_output.md

Adobe API

Usable#6 of 8

Best at keeping visual assets and financial tables in place; weaker on signatures and hierarchy.

How it scored

Complex Document Handling3/5Reading Order & Structure3/5Table Preservation4/5Visual Content Retention5/5Text & OCR Completeness4/5
Advanced Features (Bonus) · no findings1/5
Markdown Quality · no findings4/5
Complex Document Handling3/52 findings

Handles long mixed-content PDFs, but quality drops on split scanned inputs and some structural fidelity degrades in harder documents.

Struggledwhen we tried: Scanned Research Paperlink to this finding

Requires scanned PDFs above 1 MB to be split into separate files before processing, which breaks continuity across the original long document.

Struggledwhen we tried: Scanned Research Paperlink to this finding

Requires splitting one 12-page scanned paper into two PDF inputs and returns two separate markdown outputs, so the original document is not processed as a single continuous file.

adobe-api-scanned-pdf-1-6.pdf
Adobe API — Scanned PDF 1-6.pdf
adobe-pdf-extract-api-scanned-pdf-7-14.pdf
Adobe PDF Extract API — Scanned PDF 7-12.pdf
adobe-pdf-extract-api-scanned-research-pdf-pages-1-to-6-output-2.md
Loading file...
Adobe PDF Extract API — scanned_research_pdf_pages_1_to_6_output.md
adobe-pdf-extract-api-scanned-research-pdf-pages-7-to-12-output-2.md
Loading file...
Adobe PDF Extract API — scanned_research_pdf_pages_7_to_12_output.md
Reading Order & Structure3/521 findings

Document-level structure is often preserved, but TOC hierarchy, section boundaries, and some scanned-document ordering degrade.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Preserves section headings and subsection flow in a narrative financial section, keeping the main title, numbered subsection, and body text in order.

Failedwhen we tried: Scanned Research Paperlink to this finding

Section boundaries and document hierarchy are not preserved in scanned content; the title page is output without structural cues or organizational separation.

Table Preservation4/552 findings

Preserves most financial table structure היט including grouped columns and balance-sheet layouts, but breaks down on dual headers and tables interrupted by text.

Failedwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Does not distinguish row headers from column headers in dual-header tables, flattening header roles and weakening the table's structure.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Preserves compact multi-column table hierarchy and keeps header-value relationships intact in grouped tables.

Visual Content Retention5/524 findings

Charts, figures, and images are kept in place and remain visually integrated in the markdown output.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Preserves charts and images as integrated assets in the output layout instead of detaching them from the surrounding document flow.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Keeps chart elements embedded in situ and in their original position within the extracted document flow.

Text & OCR Completeness4/53 findings

Generally recovers readable text well, but misses handwritten signatures and shows some OCR/structure gaps on scanned content.

Failedwhen we tried: Target 2015 Annual Reportlink to this finding

Fails to recover handwritten signature content while still retaining the surrounding printed text, leaving the signature area incomplete.

Failedwhen we tried: Target 2015 Annual Reportlink to this finding

It does not recover handwritten signatures at all; the surrounding printed text remains, but the signature marks disappear from the parsed output.

Upstage AI

Needs work#7 of 8

Strong at native financial table reconstruction, but weak on scanned multicolumn structure and visual preservation.

Complex Document Handling3/51 finding

Handles long hybrid reports end-to-end, but quality drops on mixed-content layouts such as signatures, multicolumn text, and scanned pages.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Accepts an 84-page hybrid annual report and returns a downloadable markdown file through a fully automated API call, with no manual correction or post-processing required.

upstage-ai-upstage-hybrid-earningspdf-output-1.md
Loading file...
Upstage AI — upstage_hybrid_earningspdf_output.md
Reading Order & Structure2/532 findings

Hierarchy works in parts, but multicolumn and scanned documents lose paragraph order and section structure.

Failedwhen we tried: Target 2015 Annual Reportlink to this finding

In multicolumn narratives, headings become misaligned and the reading path diverges significantly from the source layout.

Failedwhen we tried: Scanned Research Paperlink to this finding

Loses the reading order on a dense scanned two-column page, flattening the document flow so the original section sequencing is not maintained.

Table Preservation3/527 findings

Reconstructs native financial tables well, but complex headers and some table layouts become misaligned in other documents.

Struggledwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Misaligns multi-row table headers with their data regions in complex grouped tables, producing an inconsistent table structure.

Worked wellwhen we tried: Target 2015 Annual Reportlink to this finding

Reconstructs a dense financial table with strong structural fidelity and correct value placement across columns.

Visual Content Retention2/519 findings

Charts and figures are converted into text/value extraction rather than preserved as visual assets in the right position.

Struggledwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Does not retain a waterfall chart as a visual object; it replaces the chart with a text-first analytical summary and extracted values instead of preserving the graphic in place.

Mixedwhen we tried: Target 2015 Annual Reportlink to this finding

Transforms a waterfall chart into a text-based analytical summary with extracted values and explanation, but does not retain the chart as a visual figure.

Text & OCR Completeness4/53 findings

Covers most readable content and handles scanned pages, but some currency symbols and structural details are missed.

Struggledwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Preserves the numeric amounts in the same 11-row table, but drops the currency symbols on most extracted values, so the OCR is not fully complete.

Mixedwhen we tried: Target 2015 Annual Reportlink to this finding

In reconstructed tables, it misses a small number of currency symbols, so the numeric text is not fully complete.

Advanced Features (Bonus)2/56 findings

Provides chart/table extraction and API automation, but does not clearly flag low-confidence OCR or ambiguous regions.

Mixedwhen we tried: Scanned Research Paperlink to this finding

Recovers a figure through value extraction and asset tagging, but emits raw delimiter-separated text instead of an interpretable chart or figure representation.

Mixedwhen we tried: Scanned Research Paperlink to this finding

Extracts chart values and tags the asset, but emits them as raw delimiter-separated text rather than an organized chart representation.

Nutrient.io

Needs work#8 of 8

Good at basic OCR and section hierarchy, but weak on tables, charts, and other visual content in complex documents.

How it scored

Complex Document Handling2/5Reading Order & Structure3/5Table Preservation2/5Visual Content Retention1/5
Text & OCR Completeness · no findings3/5
Advanced Features (Bonus) · no findings0/5
Markdown Quality3/5
Complex Document Handling2/53 findings

The tool processed long, mixed-content PDFs, but quality degraded on complex tables, chart pages, and scanned layouts.

Failedwhen we tried: Scanned Research Paperlink to this finding

As table complexity increases to multi-level grouped rows and column hierarchies, the parser loses the ability to preserve structural boundaries, with cells misaligned, merged incorrectly, or lost entirely.

Mixedwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Recovers some section/body hierarchy in isolated parts of an 18-page filing, but only selectively rather than consistently across the document.

Reading Order & Structure3/532 findings

Section hierarchy was preserved in some places, but paragraph flow and page-level order broke in scanned and dense documents.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

Preserves hierarchical reading order in a scanned research paper, correctly aligning section headings with the corresponding column content.

Worked wellwhen we tried: Scanned Research Paperlink to this finding

The extractor correctly aligns section headings with their corresponding column content in a multi-column scanned layout, preserving local reading order and hierarchy.

Table Preservation2/530 findings

Simple and grouped tables were sometimes usable, but multi-level headers and complex row/column relationships were often misaligned or lost.

Struggledwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Struggles to preserve a standard financial table's row-and-column structure, with significant misalignment that weakens relationships between cells and values.

Failedwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

The tool loses portions of a multi-level table header, so parent-child column relationships are not reconstructed as they appear in the source document.

Visual Content Retention1/527 findings

Chart values were extracted in linear form, but figures, chart semantics, and handwritten signatures were not retained as visual content.

Mixedwhen we tried: Scanned Research Paperlink to this finding

The extractor can recover chart values, but it corrupts the chart’s layout and does not preserve the original structured relationships, so the graphic is not retained as a chart.

Failedwhen we tried: Target 2015 Annual Reportlink to this finding

Drops the embedded portrait/image from the extracted markdown while retaining the surrounding text, so page-level visual content is not carried through.

Markdown Quality3/51 finding

Outputs were delivered as usable markdown, but structural issues and fragmented content reduced overall markdown cleanliness.

Worked wellwhen we tried: Sumitomo Heavy Industries Consolidated Financial Reportlink to this finding

Returns the extraction as a downloadable Markdown file, providing a usable markdown output format for the parsed report.

nutrient-io-nutrient-hybrid-earningspdf-output-2.md
Loading file...
Nutrient.io — nutrient_hybrid_earningspdf_output.md

Final Take

Overall, Extend AI is the best balanced pick from these scorecards. It has the strongest markdown quality (5/5), very strong reading-order structure (4.5/5), strong OCR (4.5/5), and solid complex-document handling (4/5). The main trade-off is that visual-content-retention is only mid-pack (3/5), so it is not the best option when preserving page layout and visual assets is the priority. If layout fidelity matters most, Adobe API wins that lane: it has the best visual-content-retention (5/5) and strong table preservation (4/5), with good OCR (4/5) and markdown quality (4/5). Its weaker point is hierarchy/signature handling, so it is better for visually faithful extraction than for clean semantic structure. For table-heavy documents, Landing AI is one of the top choices with table-preservation at 4/5 and solid OCR/markdown (4/5 each), but its visual retention is very weak (1/5) and reading order is only moderate (3/5). LlamaParse and Tensorlake are better if you care more about structure and reading order in mixed PDFs: both reach 4/5 on reading order, and LlamaParse is specifically strong on mixed PDFs, though it loses more on visual retention. Tensorlake is the more structured of the two, but it is still weaker on hierarchical scanned tables. Mistral AI is the best compromise when you want good OCR, better visual retention than most competitors, and export automation, but its hierarchy gets less consistent on longer documents. Upstage AI is a niche pick for native financial table reconstruction, while PDF Vector and PDF.ai are not competitive here, with PDF.ai failing outright.

Similar Tools

More tools like the ones we tested here — each links to its own tested review.

Comments (0)

Please Log in to join the discussion.