developer-tools

LlamaParse Review: PDF-to-Markdown Pipeline Tested (2026)

Name: LlamaParse
Availability: InStock
Author: AI Demos

Versatile PDF parsing for Markdown and structured JSON, with strong recovery but some fidelity drift

Visit LlamaParse

Resume JSONMulti-column PDFsField driftFree plan

TL;DR — our verdictUpdated August 2026 · 53 test artifacts

Our Take

Where it wins

You need a hosted API that converts complex PDFs into usable Markdown without manual cleanup
You work with hybrid documents that mix native text, scanned pages, tables, charts, and other visual elements
You need heading hierarchy and reading order to stay recognizable in the extracted output

Main limitation

You need perfect visual fidelity for charts, logos, signatures, or stamps instead of textual or table-based reconstructions

Pricing (verified plans)

Free $0/monthStarter $50/monthPro $500/monthEnterprise Custom pricing

Strongest test artifacts

Hosted Parsing Workspace and Reviewable JSON Export →JSON tree view →Line-item tree view →

Feature scores on this page: 10.0/10 (4 scored features)

Our take

Across the research, LlamaParse was strong at turning complex PDFs into usable downstream formats: it preserved reading order in Markdown, recovered scanned content, and produced rich nested JSON for invoices, bank statements, and resumes. The trade-off was consistency and exact structure: grouped table headers and TOC nesting could flatten, some extracted rows were incomplete or off by count, and resume field names or missing sections could drift across parses. It looks best when you want a hosted, schema-aware parsing workflow and can validate outputs before production use.

Demos by use case

Screen recording of the LlamaParse web app showing the upload-and-parse workflow on resume PDFs.

In-Depth Review

Our detailed analysis of LlamaParse — features, performance, and real-world testing.

AI Demos Team

Expert Reviewer

Verified Review

Feature-by-Feature Breakdown

Hosted Parsing Workspace and Reviewable JSON Export

Useful hosted workflow for API-backed parsing.

▾

Test Summary

Feature tested: Hosted Parsing Workspace and Reviewable JSON Export

Result: Passed — Useful hosted workflow for API-backed parsing.

Feature tested: Hosted Parsing Workspace and Reviewable JSON Export

Result: Passed

Verdict: Useful hosted workflow for API-backed parsing.

Expected behavior: LlamaParse provides a hosted web workspace for parsing and extraction, with reviewable structured JSON and confidence scores before download. The evidence mentions the Parse, Extract, Split, Classify, Sheets, and Agents UI surfaces plus downloadable outputs.

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The hosted UI shows the uploaded PDF, page controls, and the Parse/Extract workflow in the cloud app. — image-14.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The hosted UI shows the uploaded PDF, page controls, and the Parse/Extract workflow in the cloud app. — image-14.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The same cloud interface also handled the multi-column resume upload and preview. — image-15.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The same cloud interface also handled the multi-column resume upload and preview. — image-15.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Text prompt

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Text prompt): Output

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Text prompt): Output

What changed: Text prompt transformed into Text prompt

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Source bank-statement PDF used to produce downloadable JSON output. — Bank Statement PDF.pdf

Observed output: Output artifact (Text/code file): The bank statement was exported as structured JSON aligned to the supplied schema. — llamaparse-bank-statement-output.json

Input artifact: Input artifact (PDF document): Source bank-statement PDF used to produce downloadable JSON output. — Bank Statement PDF.pdf

Output artifact: Output artifact (Text/code file): The bank statement was exported as structured JSON aligned to the supplied schema. — llamaparse-bank-statement-output.json

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Source invoice PDF used to produce downloadable JSON output. — Invoice PDF.pdf

Observed output: Output artifact (Text/code file): The invoice was exported as structured JSON aligned to the supplied schema. — llamaparse-invoice-output.json

Input artifact: Input artifact (PDF document): Source invoice PDF used to produce downloadable JSON output. — Invoice PDF.pdf

Output artifact: Output artifact (Text/code file): The invoice was exported as structured JSON aligned to the supplied schema. — llamaparse-invoice-output.json

What changed: PDF document transformed into Text/code file

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Review step

Observed output: Output artifact (Image): The bank-statement output is visible as a structured JSON tree in the UI, with document nodes ordered under data > statement. — llamaparse-bank-statement-doc-order.png

Input artifact: Input artifact (Text prompt): Review step

Output artifact: Output artifact (Image): The bank-statement output is visible as a structured JSON tree in the UI, with document nodes ordered under data > statement. — llamaparse-bank-statement-doc-order.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Review step

Observed output: Output artifact (Image): The invoice output is visible in the editor/tree UI with the line_items array expanded, showing reviewable structured extraction before download. — llamaparse-invoice-extracted-line-items.png

Input artifact: Input artifact (Text prompt): Review step

Output artifact: Output artifact (Image): The invoice output is visible in the editor/tree UI with the line_items array expanded, showing reviewable structured extraction before download. — llamaparse-invoice-extracted-line-items.png

What changed: Text prompt transformed into Image

Why it matters / Conclusion: Good fit for teams that want a hosted parsing workflow and export options; the bigger buyer risk is output consistency, not upload handling.

LlamaParse provides a hosted web workspace for parsing and extraction, with reviewable structured JSON and confidence scores before download. The evidence mentions the Parse, Extract, Split, Classify, Sheets, and Agents UI surfaces plus downloadable outputs.

INPUT

Open the LlamaParse hosted app and upload a resume PDF on the Parse page.

↓→

OUTPUT

The hosted UI shows the uploaded PDF, page controls, and the Parse/Extract workflow in the cloud app.

INPUT

Open the LlamaParse hosted app and upload another resume PDF.

↓→

OUTPUT

The same cloud interface also handled the multi-column resume upload and preview.

INPUT

Check the export options reported by the research for a parsed resume.

↓→

OUTPUT

JSON, Markdown, and Excel export are reported as available.

INPUT

Bank Statement PDF.pdf

Source bank-statement PDF used to produce downloadable JSON output.

↓→

OUTPUT

llamaparse-bank-statement-output.json

Loading file...

The bank statement was exported as structured JSON aligned to the supplied schema.

INPUT

Invoice PDF.pdf

Source invoice PDF used to produce downloadable JSON output.

↓→

OUTPUT

llamaparse-invoice-output.json

Loading file...

The invoice was exported as structured JSON aligned to the supplied schema.

text

Inspect the extracted structured data in the web UI before export.

↓→

image

Output artifact for "Hosted Parsing Workspace and Reviewable JSON Export" test: The bank-statement output is visible as a structured JSON tree in the UI, with document nodes ordered under data > statement., llamaparse-bank-statement-doc-order.png

The bank-statement output is visible as a structured JSON tree in the UI, with document nodes ordered under data > statement.

text

Inspect the extracted structured data in the web UI before export.

↓→

image

Output artifact for "Hosted Parsing Workspace and Reviewable JSON Export" test: The invoice output is visible in the editor/tree UI with the line_items array expanded, showing reviewable structured extraction before download., llamaparse-invoice-extracted-line-items.png

The invoice output is visible in the editor/tree UI with the line_items array expanded, showing reviewable structured extraction before download.

Bottom Line

Good fit for teams that want a hosted parsing workflow and export options; the bigger buyer risk is output consistency, not upload handling.

From our researchParse resumes into structured data using an APIExtract and query structured data from documents using natural language

API Access and Automated Parsing

The product is set up as a hosted service with API key management and a cloud results workflow.

▾

Test Summary

Feature tested: API Access and Automated Parsing

Result: Passed — The product is set up as a hosted service with API key management and a cloud results workflow.

LlamaParse exposes project API keys and supports automated parsing through API calls, backed by a cloud results dashboard. The evidence points to API-driven workflows rather than only the web UI.

INPUT

Hosted parse workflow with API keys and downloadable results

↓→

OUTPUT

The settings page shows Project API Keys, existing keys, and a Generate New Key control for the hosted service.

INPUT

Post-run cloud results interface for parsed documents

↓→

OUTPUT

The web app shows Docs, Configs, History, Parse/Extract/Split/Classify navigation plus source-page thumbnails and result controls, confirming a hosted workflow.

Bottom Line

Well suited to API-backed pipelines, with visible credential management and a cloud results interface.

From our researchConvert a Complex PDF into Clean Markdown with an API

Table Extraction and Row Reconstruction

▾

Test Summary

Feature tested: Table Extraction and Row Reconstruction

Result: Partial

Feature tested: Table Extraction and Row Reconstruction

Result: Partial

Expected behavior: LlamaParse reconstructs tables and row records from PDFs into markdown-style tables or separate records. The examples include bank-statement transactions, invoice line items, financial tables, multi-level segment tables, and nested stand-data tables.

Test case: PDF document → Image

Input type: PDF document

Input used: Input artifact (PDF document): Input PDF — Bank Statement PDF.pdf

Observed output: Output artifact (Image): A single bank-statement transaction was extracted as a structured row with date, description, deposit amount, withdrawal amount, and running balance. — llamaparse-bank-statement-transaction-item.png

Input artifact: Input artifact (PDF document): Input PDF — Bank Statement PDF.pdf

Output artifact: Output artifact (Image): A single bank-statement transaction was extracted as a structured row with date, description, deposit amount, withdrawal amount, and running balance. — llamaparse-bank-statement-transaction-item.png

What changed: PDF document transformed into Image

Test case: PDF document → Image

Input type: PDF document

Input used: Input artifact (PDF document): Input PDF — Bank Statement PDF.pdf

Observed output: Output artifact (Image): The transactions tree reached item 54, which matches the report’s note that the extracted array was longer than the source statement’s 51 records. — llamaparse-bank-statement-transaction-count.png

Input artifact: Input artifact (PDF document): Input PDF — Bank Statement PDF.pdf

Output artifact: Output artifact (Image): The transactions tree reached item 54, which matches the report’s note that the extracted array was longer than the source statement’s 51 records. — llamaparse-bank-statement-transaction-count.png

What changed: PDF document transformed into Image

Test case: PDF document → Image

Input type: PDF document

Input used: Input artifact (PDF document): Input PDF — Invoice PDF.pdf

Observed output: Output artifact (Image): An invoice line item was extracted with line_number, channel, description, time slot, air date, rate, air time, and flight-period fields. — llamaparse-invoice-extracted-line-item-2.png

Input artifact: Input artifact (PDF document): Input PDF — Invoice PDF.pdf

Output artifact: Output artifact (Image): An invoice line item was extracted with line_number, channel, description, time slot, air date, rate, air time, and flight-period fields. — llamaparse-invoice-extracted-line-item-2.png

What changed: PDF document transformed into Image

Test case: PDF document → Image

Input type: PDF document

Input used: Input artifact (PDF document): Input PDF — Invoice PDF.pdf

Observed output: Output artifact (Image): The final visible invoice row begins with line_number 9 even though the source invoice contained eight advertising records, showing a numbering mismatch. — llamaparse-invoice-extracted-line-item-8.png

Input artifact: Input artifact (PDF document): Input PDF — Invoice PDF.pdf

Output artifact: Output artifact (Image): The final visible invoice row begins with line_number 9 even though the source invoice contained eight advertising records, showing a numbering mismatch. — llamaparse-invoice-extracted-line-item-8.png

What changed: PDF document transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — earnings_hybridInput_table.png

Observed output: Output artifact (Image): Preserved the financial summary table's rows, columns, and yearly values in a readable table. — Llamaparse_hybridInput_table_retention.png

Input artifact: Input artifact (Image): Input — earnings_hybridInput_table.png

Output artifact: Output artifact (Image): Preserved the financial summary table's rows, columns, and yearly values in a readable table. — Llamaparse_hybridInput_table_retention.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — financial_pdf_multilevel_table.png

Observed output: Output artifact (Image): Preserved the grouped segment-results table and its year-over-year columns. — financialpdf_parsed_multilevel_table.png

Input artifact: Input artifact (Image): Input — financial_pdf_multilevel_table.png

Output artifact: Output artifact (Image): Preserved the grouped segment-results table and its year-over-year columns. — financialpdf_parsed_multilevel_table.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — scanned_pdf_multicolumn_table.png

Observed output: Output artifact (Image): Cleaned up the scanned harvest-diameter table while keeping treatment rows and before/after values aligned. — llamaparse_scannedInput_table_retention.png

Input artifact: Input artifact (Image): Input — scanned_pdf_multicolumn_table.png

Output artifact: Output artifact (Image): Cleaned up the scanned harvest-diameter table while keeping treatment rows and before/after values aligned. — llamaparse_scannedInput_table_retention.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — scanned_pdf_nested_table.png

Observed output: Output artifact (Image): Reconstructed the nested stand-data table with grouped treatment columns for 7-inch through clearcut cuts. — llamaparse_scannedInput_nested_table_outputmd.png

Input artifact: Input artifact (Image): Input — scanned_pdf_nested_table.png

Output artifact: Output artifact (Image): Reconstructed the nested stand-data table with grouped treatment columns for 7-inch through clearcut cuts. — llamaparse_scannedInput_nested_table_outputmd.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — financialInput_complex_table.png

Observed output: Output artifact (Image): Preserved the segment sales table values, but the grouped column semantics became less explicit in the rebuilt version. — llamaparse_financialInput_parsed_table.png

Input artifact: Input artifact (Image): Input — financialInput_complex_table.png

Output artifact: Output artifact (Image): Preserved the segment sales table values, but the grouped column semantics became less explicit in the rebuilt version. — llamaparse_financialInput_parsed_table.png

What changed: Image transformed into Image

Why it matters / Conclusion: The tool is good at separating rows, but it is not fully reliable on record boundaries or row counts: the bank statement array was over-segmented and the invoice produced an extra line-item index.

LlamaParse reconstructs tables and row records from PDFs into markdown-style tables or separate records. The examples include bank-statement transactions, invoice line items, financial tables, multi-level segment tables, and nested stand-data tables.

file

Bank Statement PDF.pdf

↓→

image

Output artifact for "Table Extraction and Row Reconstruction" test: A single bank-statement transaction was extracted as a structured row with date, description, deposit amount, withdrawal amount, and running balance., llamaparse-bank-statement-transaction-item.png

A single bank-statement transaction was extracted as a structured row with date, description, deposit amount, withdrawal amount, and running balance.

file

Bank Statement PDF.pdf

↓→

image

Output artifact for "Table Extraction and Row Reconstruction" test: The transactions tree reached item 54, which matches the report’s note that the extracted array was longer than the source statement’s 51 records., llamaparse-bank-statement-transaction-count.png

The transactions tree reached item 54, which matches the report’s note that the extracted array was longer than the source statement’s 51 records.

file

Invoice PDF.pdf

↓→

image

Output artifact for "Table Extraction and Row Reconstruction" test: An invoice line item was extracted with line_number, channel, description, time slot, air date, rate, air time, and flight-period fields., llamaparse-invoice-extracted-line-item-2.png

An invoice line item was extracted with line_number, channel, description, time slot, air date, rate, air time, and flight-period fields.

file

Invoice PDF.pdf

↓→

image

Output artifact for "Table Extraction and Row Reconstruction" test: The final visible invoice row begins with line_number 9 even though the source invoice contained eight advertising records, showing a numbering mismatch., llamaparse-invoice-extracted-line-item-8.png

The final visible invoice row begins with line_number 9 even though the source invoice contained eight advertising records, showing a numbering mismatch.

INPUT

↓→

OUTPUT

Preserved the financial summary table's rows, columns, and yearly values in a readable table.

INPUT

↓→

OUTPUT

Preserved the grouped segment-results table and its year-over-year columns.

INPUT

↓→

OUTPUT

Cleaned up the scanned harvest-diameter table while keeping treatment rows and before/after values aligned.

INPUT

↓→

OUTPUT

Reconstructed the nested stand-data table with grouped treatment columns for 7-inch through clearcut cuts.

INPUT

↓→

OUTPUT

Preserved the segment sales table values, but the grouped column semantics became less explicit in the rebuilt version.

Bottom Line

The tool is good at separating rows, but it is not fully reliable on record boundaries or row counts: the bank statement array was over-segmented and the invoice produced an extra line-item index.

From our researchExtract and query structured data from documents using natural languageConvert a Complex PDF into Clean Markdown with an API

PDF-to-Markdown Conversion

Accepted all three complex PDFs and returned markdown exports without manual cleanup.

▾

Test Summary

Feature tested: PDF-to-Markdown Conversion

Result: Partial — Accepted all three complex PDFs and returned markdown exports without manual cleanup.

Feature tested: PDF-to-Markdown Conversion

Result: Partial

Verdict: Accepted all three complex PDFs and returned markdown exports without manual cleanup.

Expected behavior: LlamaParse converts mixed digital and scanned PDFs into downloadable Markdown while preserving readable order and headings. It was exercised on a hybrid earnings report, a table-heavy financial report, and a scanned research paper.

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Input — Hybrid-Earnings-PDF.pdf

Observed output: Output artifact (Text/code file): Accepted the 84-page hybrid annual report with native text, tables, charts, and a scanned signature page, and returned a markdown export. — llamaparse_target_earnings_output.md

Input artifact: Input artifact (PDF document): Input — Hybrid-Earnings-PDF.pdf

Output artifact: Output artifact (Text/code file): Accepted the 84-page hybrid annual report with native text, tables, charts, and a scanned signature page, and returned a markdown export. — llamaparse_target_earnings_output.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Input — Sumitomo Financial PDF.pdf

Observed output: Output artifact (Text/code file): Accepted the table-heavy Sumitomo financial report and returned a markdown export. — llamaparse_financial_pdf_output.md

Input artifact: Input artifact (PDF document): Input — Sumitomo Financial PDF.pdf

Output artifact: Output artifact (Text/code file): Accepted the table-heavy Sumitomo financial report and returned a markdown export. — llamaparse_financial_pdf_output.md

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): Input — Scanned Research PDF.pdf

Observed output: Output artifact (Text/code file): Accepted the scanned research paper with multi-column text, tables, and charts and returned a markdown export. — llamaparse_scanned_pdf_output.md

Input artifact: Input artifact (PDF document): Input — Scanned Research PDF.pdf

Output artifact: Output artifact (Text/code file): Accepted the scanned research paper with multi-column text, tables, and charts and returned a markdown export. — llamaparse_scanned_pdf_output.md

What changed: PDF document transformed into Text/code file

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — earnings_hybrid_pdf_input_page_3.png

Observed output: Output artifact (Image): Retained the Target annual report page's heading, paragraph, and bullet order as a readable text hierarchy instead of flattening it. — llamaparse_hybridInput_hierarchy.png

Input artifact: Input artifact (Image): Input — earnings_hybrid_pdf_input_page_3.png

Output artifact: Output artifact (Image): Retained the Target annual report page's heading, paragraph, and bullet order as a readable text hierarchy instead of flattening it. — llamaparse_hybridInput_hierarchy.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — financialpdf_title_page.png

Observed output: Output artifact (Image): Preserved the Sumitomo report's title, disclaimer, and section ordering in a single reading flow. — llamaparse_financialInput_hierarchy.png

Input artifact: Input artifact (Image): Input — financialpdf_title_page.png

Output artifact: Output artifact (Image): Preserved the Sumitomo report's title, disclaimer, and section ordering in a single reading flow. — llamaparse_financialInput_hierarchy.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — scanned_pdf_multicolumn_section.png

Observed output: Output artifact (Image): Turned a scanned two-column page into coherent single-column prose with the section flow intact. — llamaparse_scannedInput_hierarchy.png

Input artifact: Input artifact (Image): Input — scanned_pdf_multicolumn_section.png

Output artifact: Output artifact (Image): Turned a scanned two-column page into coherent single-column prose with the section flow intact. — llamaparse_scannedInput_hierarchy.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — earnings_hybridInput_table.png

Observed output: Output artifact (Image): LlamaParse preserves the structure of the standard financial table, maintaining row alignment, column organization, and value associations in the generated Markdown. — Llamaparse_hybridInput_table_retention.png

Input artifact: Input artifact (Image): Input — earnings_hybridInput_table.png

Output artifact: Output artifact (Image): LlamaParse preserves the structure of the standard financial table, maintaining row alignment, column organization, and value associations in the generated Markdown. — Llamaparse_hybridInput_table_retention.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — financial_pdf_multilevel_table.png

Observed output: Output artifact (Image): The parser retains the multi-level column organization of this table, preserving the relationships between grouped headers and their associated data. — financialpdf_parsed_multilevel_table.png

Input artifact: Input artifact (Image): Input — financial_pdf_multilevel_table.png

Output artifact: Output artifact (Image): The parser retains the multi-level column organization of this table, preserving the relationships between grouped headers and their associated data. — financialpdf_parsed_multilevel_table.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — financialInput_complex_table.png

Observed output: Output artifact (Image): The extracted table preserves the visible values but only partially retains the semantic relationships between grouped headers and their associated columns, so parent-child column roles are less explicit. — llamaparse_financialInput_parsed_table.png

Input artifact: Input artifact (Image): Input — financialInput_complex_table.png

Output artifact: Output artifact (Image): The extracted table preserves the visible values but only partially retains the semantic relationships between grouped headers and their associated columns, so parent-child column roles are less explicit. — llamaparse_financialInput_parsed_table.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — scanned_pdf_multilevel_table.png

Observed output: Output artifact (Image): The grouped header structure is not faithfully reconstructed; the table contents remain largely recoverable, but the original organization is only partially preserved. — llamaparse_scannedInput_multilevel_table_parsed_output.png

Input artifact: Input artifact (Image): Input — scanned_pdf_multilevel_table.png

Output artifact: Output artifact (Image): The grouped header structure is not faithfully reconstructed; the table contents remain largely recoverable, but the original organization is only partially preserved. — llamaparse_scannedInput_multilevel_table_parsed_output.png

What changed: Image transformed into Image

Why it matters / Conclusion: Strong at ingesting mixed PDF types end-to-end; the tool consistently produced a usable markdown result.

LlamaParse converts mixed digital and scanned PDFs into downloadable Markdown while preserving readable order and headings. It was exercised on a hybrid earnings report, a table-heavy financial report, and a scanned research paper.

INPUT

Hybrid-Earnings-PDF.pdf

↓→

OUTPUT

llamaparse_target_earnings_output.md

Loading file...

Accepted the 84-page hybrid annual report with native text, tables, charts, and a scanned signature page, and returned a markdown export.

INPUT

Sumitomo Financial PDF.pdf

↓→

OUTPUT

llamaparse_financial_pdf_output.md

Loading file...

Accepted the table-heavy Sumitomo financial report and returned a markdown export.

INPUT

Scanned Research PDF.pdf

↓→

OUTPUT

llamaparse_scanned_pdf_output.md

Loading file...

Accepted the scanned research paper with multi-column text, tables, and charts and returned a markdown export.

INPUT

↓→

OUTPUT

Retained the Target annual report page's heading, paragraph, and bullet order as a readable text hierarchy instead of flattening it.

INPUT

↓→

OUTPUT

Preserved the Sumitomo report's title, disclaimer, and section ordering in a single reading flow.

INPUT

↓→

OUTPUT

Turned a scanned two-column page into coherent single-column prose with the section flow intact.

INPUT

↓→

OUTPUT

LlamaParse preserves the structure of the standard financial table, maintaining row alignment, column organization, and value associations in the generated Markdown.

INPUT

↓→

OUTPUT

The parser retains the multi-level column organization of this table, preserving the relationships between grouped headers and their associated data.

INPUT

↓→

OUTPUT

The extracted table preserves the visible values but only partially retains the semantic relationships between grouped headers and their associated columns, so parent-child column roles are less explicit.

INPUT

↓→

OUTPUT

The grouped header structure is not faithfully reconstructed; the table contents remain largely recoverable, but the original organization is only partially preserved.

Bottom Line

Strong at ingesting mixed PDF types end-to-end; the tool consistently produced a usable markdown result.

From our researchConvert a Complex PDF into Clean Markdown with an API

OCR, Reading Order, and Visual Element Transcription

Recovers text from scans and transcribes charts/signatures, but does not keep visuals as visuals.

▾

Test Summary

Feature tested: OCR, Reading Order, and Visual Element Transcription

Result: Partial — Recovers text from scans and transcribes charts/signatures, but does not keep visuals as visuals.

Feature tested: OCR, Reading Order, and Visual Element Transcription

Result: Partial

Verdict: Recovers text from scans and transcribes charts/signatures, but does not keep visuals as visuals.

Expected behavior: LlamaParse transcribes scanned prose and visual page elements into readable text or structured representations. The evidence covers scanned research-paper pages, multi-column reading-order recovery, charts, blurry signatures, stamps, and logos.

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — hybridearnings_pdf_waterfall_chart.png

Observed output: Output artifact (Image): The chart was translated into a text table listing the SG&A rate values and contributing changes in sequence. — llamaparse_hybrid_earningspdf_parsed_waterfall_chart.png

Input artifact: Input artifact (Image): Input — hybridearnings_pdf_waterfall_chart.png

Output artifact: Output artifact (Image): The chart was translated into a text table listing the SG&A rate values and contributing changes in sequence. — llamaparse_hybrid_earningspdf_parsed_waterfall_chart.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — hybrid_earningspdf_blurry_stamp.png

Observed output: Output artifact (Image): The blurry Ernst & Young signature mark remained recognizable in the parsed output as a detected signature line and page number. — llamaparse_hybrid_earningspdf_parsed_blurry_stamp.png

Input artifact: Input artifact (Image): Input — hybrid_earningspdf_blurry_stamp.png

Output artifact: Output artifact (Image): The blurry Ernst & Young signature mark remained recognizable in the parsed output as a detected signature line and page number. — llamaparse_hybrid_earningspdf_parsed_blurry_stamp.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Input — scanned_pdf_chart.png

Observed output: Output artifact (Image): The chart was converted into a structured table while maintaining the mapping between legend entries and their associated values. — llamaparse_scannedInput_parsed_chart.png

Input artifact: Input artifact (Image): Input — scanned_pdf_chart.png

Output artifact: Output artifact (Image): The chart was converted into a structured table while maintaining the mapping between legend entries and their associated values. — llamaparse_scannedInput_parsed_chart.png

What changed: Image transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The parser successfully reconstructs the scanned multi-column layout into a coherent reading flow, preserving the relationship between section headings and the paragraphs that follow. — llamaparse_scannedInput_hierarchy.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The parser successfully reconstructs the scanned multi-column layout into a coherent reading flow, preserving the relationship between section headings and the paragraphs that follow. — llamaparse_scannedInput_hierarchy.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The parser retains the organizational structure of the multi-column earnings report, preserving section hierarchy and content flow rather than flattening the document into disconnected text blocks. — llamaparse_hybridInput_hierarchy.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The parser retains the organizational structure of the multi-column earnings report, preserving section hierarchy and content flow rather than flattening the document into disconnected text blocks. — llamaparse_hybridInput_hierarchy.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Text prompt

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Text prompt): Output

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Text prompt): Output

What changed: Text prompt transformed into Text prompt

Why it matters / Conclusion: OCR and transcription coverage is good, but the output is text-centric rather than image-preserving.

LlamaParse transcribes scanned prose and visual page elements into readable text or structured representations. The evidence covers scanned research-paper pages, multi-column reading-order recovery, charts, blurry signatures, stamps, and logos.

INPUT

↓→

OUTPUT

The chart was translated into a text table listing the SG&A rate values and contributing changes in sequence.

INPUT

↓→

OUTPUT

The blurry Ernst & Young signature mark remained recognizable in the parsed output as a detected signature line and page number.

INPUT

↓→

OUTPUT

The chart was converted into a structured table while maintaining the mapping between legend entries and their associated values.

INPUT

INPUT: A scanned research paper page headed "STUDY AREA" with dense two-column prose and a following section heading "STAND PRESCRIPTIONS."

↓→

OUTPUT

The parser successfully reconstructs the scanned multi-column layout into a coherent reading flow, preserving the relationship between section headings and the paragraphs that follow.

INPUT

INPUT: Hybrid Target annual report page 3 with the heading "A Growth Story Again," two-column narrative text, and bullet points.

↓→

OUTPUT

The parser retains the organizational structure of the multi-column earnings report, preserving section hierarchy and content flow rather than flattening the document into disconnected text blocks.

INPUT

INPUT: Target logo placeholder and Brian Cornell signature block from the annual report.

↓→

OUTPUT

Rather than omitting non-text assets, the extracted content describes logos and signatures within the output text.

Bottom Line

OCR and transcription coverage is good, but the output is text-centric rather than image-preserving.

From our researchConvert a Complex PDF into Clean Markdown with an API

Schema-driven resume extraction

Strong

▾

Test Summary

Feature tested: Schema-driven resume extraction

Result: Partial — Strong

Feature tested: Schema-driven resume extraction

Result: Partial

Verdict: Strong

Expected behavior: LlamaParse can ingest uploaded resume PDFs and return structured JSON from a defined extraction schema, as exercised on clean single-column, two-column, and dense messy resumes. It also captured a broad set of resume sections and normalized them into exportable structured output, though the exact key structure varied across documents.

Test case: Image → Text prompt

Input type: Image

Input used: Input artifact (Image): Clean single-column resume PDF for Rugved Nichite uploaded in the Parse view with a custom extraction schema. — image-14.png

Observed output: Output artifact (Text prompt): Parsed result

Input artifact: Input artifact (Image): Clean single-column resume PDF for Rugved Nichite uploaded in the Parse view with a custom extraction schema. — image-14.png

Output artifact: Output artifact (Text prompt): Parsed result

What changed: Image transformed into Text prompt

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Multi-column resume PDF for Priya Sharma. — image-15.png

Observed output: Output artifact (Image): Certification issuer data was present on the clean resume but missing here, showing inconsistent field depth across parses. — certifecation Section missing.png

Input artifact: Input artifact (Image): Multi-column resume PDF for Priya Sharma. — image-15.png

Output artifact: Output artifact (Image): Certification issuer data was present on the clean resume but missing here, showing inconsistent field depth across parses. — certifecation Section missing.png

What changed: Image transformed into Image

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Messy resume PDF for John Kumar. — image-7.png

Observed output: Output artifact (Image): The languages key was omitted entirely, '72 percent marks' was not normalized to 72%, and skills were kept in lowercase. — LlamaParse extracted the certification issuer.png

Input artifact: Input artifact (Image): Messy resume PDF for John Kumar. — image-7.png

Output artifact: Output artifact (Image): The languages key was omitted entirely, '72 percent marks' was not normalized to 72%, and skills were kept in lowercase. — LlamaParse extracted the certification issuer.png

What changed: Image transformed into Image

Test case: Image → Text prompt

Input type: Image

Input used: Input artifact (Image): Clean single-column resume PDF for Rugved Nichite. — image-14.png

Observed output: Output artifact (Text prompt): Observed parse

Input artifact: Input artifact (Image): Clean single-column resume PDF for Rugved Nichite. — image-14.png

Output artifact: Output artifact (Text prompt): Observed parse

What changed: Image transformed into Text prompt

Test case: Image → Text prompt

Input type: Image

Input used: Input artifact (Image): Clean resume PDF containing summary, work history, education, skills, and certifications. — image-14.png

Observed output: Output artifact (Text prompt): Observed output

Input artifact: Input artifact (Image): Clean resume PDF containing summary, work history, education, skills, and certifications. — image-14.png

Output artifact: Output artifact (Text prompt): Observed output

What changed: Image transformed into Text prompt

Test case: Image → Image

Input type: Image

Input used: Input artifact (Image): Clean resume PDF for Rugved Nichite. — image-14.png

Observed output: Output artifact (Image): Observed issues — Job title missing.png

Input artifact: Input artifact (Image): Clean resume PDF for Rugved Nichite. — image-14.png

Output artifact: Output artifact (Image): Observed issues — Job title missing.png

What changed: Image transformed into Image

Why it matters / Conclusion: Excellent at turning resumes into structured JSON, but the output schema is not fully deterministic across documents.

LlamaParse can ingest uploaded resume PDFs and return structured JSON from a defined extraction schema, as exercised on clean single-column, two-column, and dense messy resumes. It also captured a broad set of resume sections and normalized them into exportable structured output, though the exact key structure varied across documents.

image

Input artifact for "Schema-driven resume extraction" test: Clean single-column resume PDF for Rugved Nichite uploaded in the Parse view with a custom extraction schema., image-14.png

Clean single-column resume PDF for Rugved Nichite uploaded in the Parse view with a custom extraction schema.

↓→

RESULT

Parsed successfully into structured JSON. Contact details, work history, education, skills, and certifications were extracted; skills came back as categorized arrays, certifications as structured objects, and responsibilities as separate array items. The job title dropped the leading 'AI', and the education CGPA appeared as 'CGPA: 8.2 / 10'.

image

Multi-column resume PDF for Priya Sharma.

↓→

text

Output artifact for "Schema-driven resume extraction" test: Certification issuer data was present on the clean resume but missing here, showing inconsistent field depth across parses., certifecation Section missing.png

Certification issuer data was present on the clean resume but missing here, showing inconsistent field depth across parses.

image

Messy resume PDF for John Kumar.

↓→

text

Output artifact for "Schema-driven resume extraction" test: The languages key was omitted entirely, '72 percent marks' was not normalized to 72%, and skills were kept in lowercase., LlamaParse extracted the certification issuer.png

The languages key was omitted entirely, '72 percent marks' was not normalized to 72%, and skills were kept in lowercase.

image

Clean single-column resume PDF for Rugved Nichite.

↓→

RESULT

Parsed successfully in the web app with the resume rendered in the viewer and a structured result produced from the clean layout.

image

Input artifact for "Schema-driven resume extraction" test: Clean resume PDF containing summary, work history, education, skills, and certifications., image-14.png

Clean resume PDF containing summary, work history, education, skills, and certifications.

↓→

RESULT

Returned contact details, a professional summary, two work entries with responsibilities, education, categorized skills, and structured certifications. The output was the most structurally rich of the clean-resume tests.

image

Clean resume PDF for Rugved Nichite.

↓→

RESULT

Output artifact for "Schema-driven resume extraction" test: Observed issues, Job title missing.png

Bottom Line

Excellent at turning resumes into structured JSON, but the output schema is not fully deterministic across documents.

From our researchParse resumes into structured data using an API

Structured Resume Parsing

Excellent — most structurally rich output of all tools tested, CGPA and certifications fully structured

10/10

▾

Test Summary

Feature tested: Structured Resume Parsing

Result: Passed (10/10) — Excellent — most structurally rich output of all tools tested, CGPA and certifications fully structured

Feature tested: Structured Resume Parsing

Result: Passed (10/10)

Verdict: Excellent — most structurally rich output of all tools tested, CGPA and certifications fully structured

Expected behavior: Extracts resume fields into structured JSON across clean, multi-column, and messy layouts. In the tests it handled a standard single-column resume, a multi-column resume without layout hints, and a highly inconsistent resume with missing headers and comma-separated lists.

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): input-1-clean-resume-rugved.pdf — Llamaparse input.1.pdf

Observed output: Output artifact (Text/code file): Full JSON output — LlamaParse parsing clean resume — llama output.1.txt

Input artifact: Input artifact (PDF document): input-1-clean-resume-rugved.pdf — Llamaparse input.1.pdf

Output artifact: Output artifact (Text/code file): Full JSON output — LlamaParse parsing clean resume — llama output.1.txt

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): nput-2-multicolumn-resume-priya.pdf — Llamaparse input.2.pdf

Observed output: Output artifact (Text/code file): Full JSON output — LlamaParse parsing multi-column resume — llama output.2.txt

Input artifact: Input artifact (PDF document): nput-2-multicolumn-resume-priya.pdf — Llamaparse input.2.pdf

Output artifact: Output artifact (Text/code file): Full JSON output — LlamaParse parsing multi-column resume — llama output.2.txt

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): input-3-messy-resume-john.pdf — Llamaparse input.3.pdf

Observed output: Output artifact (Text/code file): Full JSON output — LlamaParse parsing messy resume — llama output.3.txt

Input artifact: Input artifact (PDF document): input-3-messy-resume-john.pdf — Llamaparse input.3.pdf

Output artifact: Output artifact (Text/code file): Full JSON output — LlamaParse parsing messy resume — llama output.3.txt

What changed: PDF document transformed into Text/code file

Why it matters / Conclusion: Excellent output on clean resumes — most structurally rich of all tools tested. CGPA captured as dedicated standalone field. All 5 skill categories correctly structured. Both certifications as fully structured objects. Main weakness is job title missing the AI prefix and languages field absent since no spoken languages section was in the resume.

Extracts resume fields into structured JSON across clean, multi-column, and messy layouts. In the tests it handled a standard single-column resume, a multi-column resume without layout hints, and a highly inconsistent resume with missing headers and comma-separated lists.

PDF

Llamaparse input.1.pdf

↓→

PDF

llama output.1.txt

Loading file...

PDF

Llamaparse input.2.pdf

↓→

PDF

llama output.2.txt

Loading file...

PDF

Llamaparse input.3.pdf

↓→

PDF

llama output.3.txt

Loading file...

Bottom Line

Excellent output on clean resumes — most structurally rich of all tools tested. CGPA captured as dedicated standalone field. All 5 skill categories correctly structured. Both certifications as fully structured objects. Main weakness is job title missing the AI prefix and languages field absent since no spoken languages section was in the resume.

From our researchearlier research

Structured Resume Parsing

Strong output richness, but schema stability is uneven.

10/10

▾

Test Summary

Feature tested: Structured Resume Parsing

Result: Partial (10/10) — Strong output richness, but schema stability is uneven.

LlamaParse extracts machine-readable JSON from resume PDFs across clean single-column, multi-column, and messy layouts. The runs surfaced work history, education, skills, certifications, languages, projects, and other nested sections.

INPUT

Multi-column resume PDF for Priya Sharma with a right-side languages section and dedicated projects section.

↓→

OUTPUT

llama output 2.txt

Loading file...

Languages were returned with proficiency levels and projects were captured as dedicated records.

Bottom Line

Very strong extraction breadth, but the output schema is not stable enough for downstream systems that rely on fixed keys.

From our researchearlier research

Schema-Guided Structured Data Extraction

▾

Test Summary

Feature tested: Schema-Guided Structured Data Extraction

Result: Passed

LlamaParse can take a PDF plus a user-defined JSON schema, or seed a schema from the document, and populate nested structured JSON rather than flat OCR text. In the bank statement and invoice runs it filled fields like metadata, account details, balances, and invoice line items.

file

Bank Statement PDF.pdf

↓→

image

Output artifact for "Schema-Guided Structured Data Extraction" test: The bank statement extraction followed the requested hierarchy, including metadata, account holder, account, branch, statement-period, balances, transactions, rewards, and disclaimers sections., llamaparse-bank-statement-extracted-details.png

The bank statement extraction followed the requested hierarchy, including metadata, account holder, account, branch, statement-period, balances, transactions, rewards, and disclaimers sections.

file

Invoice PDF.pdf

↓→

image

Output artifact for "Schema-Guided Structured Data Extraction" test: The invoice extraction followed the supplied schema and reconstructed invoice metadata, advertiser, and station objects as structured JSON., llamaparse-invoice-metadata.png

The invoice extraction followed the supplied schema and reconstructed invoice metadata, advertiser, and station objects as structured JSON.

Bottom Line

Strong when the schema is explicit and the document has a clear financial structure, but downstream validation is still needed because some extracted fields remained blank or inconsistent in later row-level tests.

From our researchExtract and query structured data from documents using natural languageParse resumes into structured data using an API

Totals and Summary Field Extraction

▾

Test Summary

Feature tested: Totals and Summary Field Extraction

Result: Passed

LlamaParse can extract document-level totals and rollups alongside detailed rows. In the bank statement and invoice tests it surfaced figures such as total deposits, total withdrawals, gross total, net amount due, and payment terms.

INPUT

Bank Statement PDF.pdf

Source bank-statement PDF used to test summary extraction.

↓→

OUTPUT

The bank statement summary object reported total deposits, total withdrawals, and total transactions.

INPUT

Invoice PDF.pdf

Source invoice PDF used to test summary extraction.

↓→

OUTPUT

The invoice summary object reported aired spots, gross total, agency commission, net amount due, and payment terms.

Bottom Line

Financial rollups are a useful companion to row extraction, and the invoice totals matched the source cleanly; however, the bank statement transaction count in summary did not reconcile with the extracted rows or source statement.

From our researchExtract and query structured data from documents using natural language

Table of Contents Extraction

▾

Test Summary

Feature tested: Table of Contents Extraction

Result: Passed

Extracts TOC entries and page numbers from report front matter. The tested output recovered TOC content even when the full hierarchy was not always rebuilt.

INPUT

INPUT: Financial report table-of-contents page with major sections, subsections, and page numbers.

↓→

OUTPUT

The TOC entries and page numbers were recovered, but the organizational relationships that define the TOC were not preserved; the page came out as sequential text rather than a structured table.

Bottom Line

Useful for recovering TOC content, but not for preserving full TOC structure.

From our researchConvert a Complex PDF into Clean Markdown with an API

Resume Information Extraction

Excellent — most structurally rich output of all tools tested, CGPA and certifications fully structured

10/10

▾

Test Summary

Feature tested: Resume Information Extraction

Result: Passed (10/10) — Excellent — most structurally rich output of all tools tested, CGPA and certifications fully structured

Feature tested: Resume Information Extraction

Result: Passed (10/10)

Verdict: Excellent — most structurally rich output of all tools tested, CGPA and certifications fully structured

Expected behavior: LlamaParse extracts structured fields from resumes across clean single-column, multi-column, and messy formats, returning rich JSON with skills, certifications, languages, projects, education, and related fields.

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): input-1-clean-resume-rugved.pdf — Llamaparse input.1.pdf

Observed output: Output artifact (Text/code file): Full JSON output — LlamaParse parsing clean resume — llama output.1.txt

Input artifact: Input artifact (PDF document): input-1-clean-resume-rugved.pdf — Llamaparse input.1.pdf

Output artifact: Output artifact (Text/code file): Full JSON output — LlamaParse parsing clean resume — llama output.1.txt

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): nput-2-multicolumn-resume-priya.pdf — Llamaparse input.2.pdf

Observed output: Output artifact (Text/code file): Full JSON output — LlamaParse parsing multi-column resume — llama output.2.txt

Input artifact: Input artifact (PDF document): nput-2-multicolumn-resume-priya.pdf — Llamaparse input.2.pdf

Output artifact: Output artifact (Text/code file): Full JSON output — LlamaParse parsing multi-column resume — llama output.2.txt

What changed: PDF document transformed into Text/code file

Test case: PDF document → Text/code file

Input type: PDF document

Input used: Input artifact (PDF document): input-3-messy-resume-john.pdf — Llamaparse input.3.pdf

Observed output: Output artifact (Text/code file): Full JSON output — LlamaParse parsing messy resume — llama output.3.txt

Input artifact: Input artifact (PDF document): input-3-messy-resume-john.pdf — Llamaparse input.3.pdf

Output artifact: Output artifact (Text/code file): Full JSON output — LlamaParse parsing messy resume — llama output.3.txt

What changed: PDF document transformed into Text/code file

LlamaParse extracts structured fields from resumes across clean single-column, multi-column, and messy formats, returning rich JSON with skills, certifications, languages, projects, education, and related fields.

PDF

Llamaparse input.1.pdf

↓→

PDF

llama output.1.txt

Loading file...

PDF

Llamaparse input.2.pdf

↓→

PDF

llama output.2.txt

Loading file...

PDF

Llamaparse input.3.pdf

↓→

PDF

llama output.3.txt

Loading file...

Bottom Line

From our researchearlier research

Resume Parsing

Excellent — most structurally rich output of all tools tested, CGPA and certifications fully structured

10/10

▾

Test Summary

Feature tested: Resume Parsing

Result: Passed (10/10) — Excellent — most structurally rich output of all tools tested, CGPA and certifications fully structured

Feature tested: Resume Parsing

Result: Passed (10/10)

Verdict: Excellent — most structurally rich output of all tools tested, CGPA and certifications fully structured

Test case: PDF document → Image

Input type: PDF document

Input used: Input artifact (PDF document): input-1-clean-resume-rugved.pdf — Llamaparse input.1.pdf

Observed output: Output artifact (Image): Both fields are absent because the resume had no data for them — no start date in education, no languages section. However LlamaParse drops the keys entirely instead of returning null or an empty array. A downstream system expecting these keys will get a KeyError with no warning. — no start date in education, no languages section..png

Input artifact: Input artifact (PDF document): input-1-clean-resume-rugved.pdf — Llamaparse input.1.pdf

Output artifact: Output artifact (Image): Both fields are absent because the resume had no data for them — no start date in education, no languages section. However LlamaParse drops the keys entirely instead of returning null or an empty array. A downstream system expecting these keys will get a KeyError with no warning. — no start date in education, no languages section..png

What changed: PDF document transformed into Image

Test case: PDF document → Image

Input type: PDF document

Input used: Input artifact (PDF document): nput-2-multicolumn-resume-priya.pdf — Llamaparse input.2.pdf

Observed output: Output artifact (Image): LlamaParse extracted the certification issuer correctly for Input 1 but silently dropped it for Input 2. As you can see in the comparison — Input 1 returns "issuer": "Amazon Web Services" and "issuer": "IBM / Coursera" as dedicated fields, while Input 2 returns only name and year with no issuer field at all. Both resumes had certification issuer information clearly written. A recruiter verifying whether an AWS certification came from Amazon or a third-party provider would find the answer for one candidate but not another — with no error, no null field, just a completely absent key. — INCONSISTENT DEPTH for certifications.png

Input artifact: Input artifact (PDF document): nput-2-multicolumn-resume-priya.pdf — Llamaparse input.2.pdf

Output artifact: Output artifact (Image): LlamaParse extracted the certification issuer correctly for Input 1 but silently dropped it for Input 2. As you can see in the comparison — Input 1 returns "issuer": "Amazon Web Services" and "issuer": "IBM / Coursera" as dedicated fields, while Input 2 returns only name and year with no issuer field at all. Both resumes had certification issuer information clearly written. A recruiter verifying whether an AWS certification came from Amazon or a third-party provider would find the answer for one candidate but not another — with no error, no null field, just a completely absent key. — INCONSISTENT DEPTH for certifications.png

What changed: PDF document transformed into Image

Test case: PDF document → Image

Input type: PDF document

Input used: Input artifact (PDF document): input-3-messy-resume-john.pdf — Llamaparse input.3.pdf

Observed output: Output artifact (Image): LlamaParse uses GPT-based extraction which interprets the schema instructions differently for each document. As you can see in the comparison above — Input 1 returns candidate name as a top-level field, Input 2 wraps everything inside a "personal_info" block, and Input 3 returns name at top level but with a separate "contact" block. All three parse the same data but use three completely different key structures. A production system consuming resumes at scale would silently break on Input 2 while working fine on Inputs 1 and 3 — with no error thrown, just a missing name field in the output. — INCONSISTENT FIELD NAMING for all the inputs.png

Input artifact: Input artifact (PDF document): input-3-messy-resume-john.pdf — Llamaparse input.3.pdf

Output artifact: Output artifact (Image): LlamaParse uses GPT-based extraction which interprets the schema instructions differently for each document. As you can see in the comparison above — Input 1 returns candidate name as a top-level field, Input 2 wraps everything inside a "personal_info" block, and Input 3 returns name at top level but with a separate "contact" block. All three parse the same data but use three completely different key structures. A production system consuming resumes at scale would silently break on Input 2 while working fine on Inputs 1 and 3 — with no error thrown, just a missing name field in the output. — INCONSISTENT FIELD NAMING for all the inputs.png

What changed: PDF document transformed into Image

Test case: Artifact → Image

Input type: Artifact

Input used: Input artifact (Artifact): Input 1 Rugved Nichite Clean resume

Observed output: Output artifact (Image): CGPA returned as "CGPA: 8.2 / 10" — the full label and value packed into a single string in the grade_percentage field instead of a clean numeric value like 8.2. — CGPA embedded in string — not a clean numeric field.png

Input artifact: Input artifact (Artifact): Input 1 Rugved Nichite Clean resume

Output artifact: Output artifact (Image): CGPA returned as "CGPA: 8.2 / 10" — the full label and value packed into a single string in the grade_percentage field instead of a clean numeric value like 8.2. — CGPA embedded in string — not a clean numeric field.png

What changed: Artifact transformed into Image

Test case: Artifact → Image

Input type: Artifact

Input used: Input artifact (Artifact): John Kumar messy resume input 3

Observed output: Output artifact (Image): LlamaParse omits the languages key entirely when the resume has no languages section — instead of returning an empty array. As you can see — Input 2 returns a fully structured languages array with proficiency levels, while Input 3 has no languages key anywhere in the output. A downstream system expecting a languages key will throw a KeyError or null reference error, while tools like Affinda and ResumeParser handle this gracefully by returning an empty array. — ABSENT FIELD language on inputs 3.png

Input artifact: Input artifact (Artifact): John Kumar messy resume input 3

Output artifact: Output artifact (Image): LlamaParse omits the languages key entirely when the resume has no languages section — instead of returning an empty array. As you can see — Input 2 returns a fully structured languages array with proficiency levels, while Input 3 has no languages key anywhere in the output. A downstream system expecting a languages key will throw a KeyError or null reference error, while tools like Affinda and ResumeParser handle this gracefully by returning an empty array. — ABSENT FIELD language on inputs 3.png

What changed: Artifact transformed into Image

Test case: Artifact → Image

Input type: Artifact

Input used: Input artifact (Artifact): John Kumar messy resume input 3

Observed output: Output artifact (Image): Skills returned as "python", "java", "html", "css" — all lowercase as written in the source resume. No capitalisation normalisation was applied despite the skills being well-known proper nouns. — All skills returned in lowercase — original casing not preserved.png

Input artifact: Input artifact (Artifact): John Kumar messy resume input 3

Output artifact: Output artifact (Image): Skills returned as "python", "java", "html", "css" — all lowercase as written in the source resume. No capitalisation normalisation was applied despite the skills being well-known proper nouns. — All skills returned in lowercase — original casing not preserved.png

What changed: Artifact transformed into Image

PDF

Llamaparse input.1.pdf

↓→

PDF

Output artifact for "Resume Parsing" test: Both fields are absent because the resume had no data for them — no start date in education, no languages section. However LlamaParse drops the keys entirely instead of returning null or an empty array. A downstream system expecting these keys will get a KeyError with no warning., no start date in education, no languages section..png

PDF

Llamaparse input.2.pdf

↓→

PDF

Output artifact for "Resume Parsing" test: LlamaParse extracted the certification issuer correctly for Input 1 but silently dropped it for Input 2. As you can see in the comparison — Input 1 returns "issuer": "Amazon Web Services" and "issuer": "IBM / Coursera" as dedicated fields, while Input 2 returns only name and year with no issuer field at all. Both resumes had certification issuer information clearly written. A recruiter verifying whether an AWS certification came from Amazon or a third-party provider would find the answer for one candidate but not another — with no error, no null field, just a completely absent key., INCONSISTENT DEPTH for certifications.png

PDF

Llamaparse input.3.pdf

↓→

PDF

Output artifact for "Resume Parsing" test: LlamaParse uses GPT-based extraction which interprets the schema instructions differently for each document. As you can see in the comparison above — Input 1 returns candidate name as a top-level field, Input 2 wraps everything inside a "personal_info" block, and Input 3 returns name at top level but with a separate "contact" block. All three parse the same data but use three completely different key structures. A production system consuming resumes at scale would silently break on Input 2 while working fine on Inputs 1 and 3 — with no error thrown, just a missing name field in the output., INCONSISTENT FIELD NAMING for all the inputs.png

INPUT

Input 1 Rugved Nichite Clean resume

↓→

OUTPUT

INPUT

John Kumar messy resume input 3

↓→

OUTPUT

INPUT

John Kumar messy resume input 3

↓→

OUTPUT

Bottom Line

From our researchearlier research

Plans reported in the evaluation

Free

$0/month

Includes 10,000 credits per month, 1 user, and basic support.

Starter

$50/month

Includes 40,000 credits per month, pay-as-you-go usage up to 400,000 credits, and supports up to 5 users.

Pro

$500/month

Includes 400,000 credits per month, pay-as-you-go usage up to 4 million credits, supports up to 10 users, and includes Slack support.

Enterprise

Custom pricing

Includes volume discounts, higher rate limits, SSO, SaaS or hybrid deployment options, and dedicated account management.

✓ Use This If

●You need a hosted API that converts complex PDFs into usable Markdown without manual cleanup

●You work with hybrid documents that mix native text, scanned pages, tables, charts, and other visual elements

●You need heading hierarchy and reading order to stay recognizable in the extracted output

●You need schema-driven extraction into nested JSON for documents like invoices, bank statements, or resumes

●You want downloadable JSON or Markdown returned programmatically, with confidence scores available for review

✕ Skip This If

●You need perfect visual fidelity for charts, logos, signatures, or stamps instead of textual or table-based reconstructions

●You need grouped multi-level table headers to remain fully explicit in every case

●You need the table of contents reconstructed as a fully nested hierarchy rather than sequential text

●You need every extracted row to have complete IDs, transaction types, and value dates without manual validation

●You need the same field names, wrappers, and null/empty placeholders on every resume parse

Featured in Rankings

Independent rankings where LlamaParse was tested and rated.

RankingsUsable

Best AI APIs to Convert Complex PDFs to Clean Markdown

We tested hosted PDF-to-markdown APIs on the same three hard documents: a...

Rankings

RankingsUsable

Best AI Tools for Extracting Structured Data from PDFs and Business Documents

If you need to turn invoices and bank statements into clean, nested...

Rankings

RankingsUsable

Best AI Resume Parser APIs for Structured JSON Output

If you need to turn resume PDFs into clean, machine-readable JSON for...

developer-toolsapistextOther

Yes. In this research it accepted an 84-page hybrid earnings report and a scanned research paper, and the output preserved readable content and document flow instead of skipping the scanned material.

It preserved reading order and kept document structure recognizable in the Markdown output. The trade-off was that the table of contents was flattened into sequential text rather than kept as a fully nested hierarchy.

Standard tables were preserved well, and many multi-level tables stayed readable, including financial summary and segment tables. The main limitation was that grouped header semantics became less explicit in some complex tables.

Charts were not kept as visual charts in the output. Instead, they were converted into structured text or table representations, while logos, signatures, and even a blurry stamp were still represented in extracted text or detected as recognizable elements.

Yes. In this research it extracted both a bank statement and an invoice into nested JSON that followed the supplied schemas, with detailed hierarchy such as metadata, account or advertiser details, line items, summaries, and disclaimers.

Structural fidelity was not perfect at the edges. The bank statement repeatedly left transaction_id and transaction_type empty, some rows had blank value_date fields, and record counts drifted from the source; the invoice also showed a line-number mismatch against its summary.

Yes. The report says you can define a schema manually field-by-field, paste a full JSON schema, or use the AI Schema Generator, and the extracted data can be downloaded in JSON format with confidence scores shown in the interface.

It parsed clean, multi-column, and messy resume PDFs successfully in the cloud UI, and it was strongest on skills, certifications, languages, work history, education, and other detailed nested fields. The weakness was stability: field names could drift, missing sections could disappear instead of coming back as nulls or empty arrays, and some values were not normalized consistently.

Banner Preview

How the embed badge will look on your site

Embed HTML

Copy this code to your website source

Quick Integration Guide

1Copy the HTML code block above.
2Paste it into your site's HTML or CMS editor.
3Banner appears instantly on your page.
4Links back to your tool profile here.

Similar Tools

Discover more AI tools like LlamaParse to enhance your workflow.

🤖

Affinda

Best overall resume parsing API here for clean, multi-column, and messy PDFs with rich structured JSON.

AI Tool

🤖

Airparser

Parses clean, multi-column, and messy resumes into structured JSON, but email, title, and skill formatting still need validation.

AI Tool

🤖

Extracta.ai

Schema-first resume parsing that stays lean and predictable across clean, multi-column, and messy PDFs.

AI Tool

🤖

Skima AI

Fast PDF resume parsing with dependable core fields and experience-year calculation, but weak structured output and supplemental coverage.

AI Tool

🤖

Parseur

Template-driven resume parsing that returns predictable JSON after one-time setup.

AI Tool

🤖

HrFlow

HrFlow is an API-first resume parser that covers core fields well, but needs cleanup for phones, casing, and certifications.

AI Tool

🤖

OpenResume

Free browser resume parser that is handy for quick manual review on clean PDFs, but brittle field mapping and no JSON export make it unsuitable for API pipelines.

AI Tool

🤖

CVParserPro

Parses resume PDFs into structured candidate profiles quickly, but experience totals and education dates need manual review.

AI Tool

🤖

Hireability

Strong resume-to-JSON parsing on standard and messy single-column PDFs, but two-column layouts can break the result schema.

AI Tool

🤖

Docparser

Template-based resume parsing that works on one fixed layout, but breaks on varied resumes.

AI Tool

🤖

Landing AI

Schema-guided PDF extraction for bank statements and invoices, with strong row capture and a few identifier QA caveats.

A strong hosted PDF-to-markdown API for mixed and scanned documents, with solid OCR, table recovery, and asset export but uneven structural fidelity.

Solid on native financial tables, but unreliable for multi-column and scanned-document structure in markdown conversion.

Schema-driven extraction for finance PDFs that reconstructs nested JSON well, but still needs review for ordering, IDs, and a few scalar values.

AI Tool

🤖

Nanonets

Schema-first PDF extraction that produces usable exports, but dense table rows still need review.

AI Tool

🤖

Retab

Schema-first PDF extraction for finance documents that returns nested JSON with minimal setup.

AI Tool

🤖

Datalab

Schema-paste extraction for bank statements and invoices, with cited JSON output and fast-mode recovery when schemas get large.

AI Tool

🤖

Reducto

Structured PDF extraction with nested JSON, citations, and row-preserving tables; invoices worked cleanly, while bank statements still need review.

Comments (0)

Built by FutureSmart AI — the team behind AI Demos

Need a custom AI solution for this use case?

If you are looking to build a custom document extraction, PDF parsing, or structured data extraction system for your business or internal workflow, email us at contact@futuresmart.ai.

Get a custom build

Found something inaccurate or missing? We try to keep our AI research accurate and useful. If you found outdated information, an issue, or have a suggestion, email us at collaborate@aidemos.com.

LlamaParse Review: PDF-to-Markdown Pipeline Tested (2026)

Our take

In-Depth Review

Feature-by-Feature Breakdown

Plans reported in the evaluation

Featured in Rankings

Promote LlamaParse

Banner Preview

Embed HTML

Quick Integration Guide

Similar Tools

Comments (0)

Need a custom AI solution for this use case?