AI Demos Research — the structured-intelligence platform. Every verdict on these pages opens to the execution behind it.

Can Extend AI read a cleanly scanned document accurately?

Extend AI accurately recovered text from a cleanly scanned, image-only document. The one-page laboratory certificate was reproduced in Markdown with the letterhead, reference, date, six result rows, closing text, sign-off, and footer matching the page render. The results block also came back as a two-column HTML table with each label beside its value.

1 of 1 test case passed

Every test case in this scenario has a result.

Pass rate100%1 of 1 with a result
Coverage1 of 1test cases with a result
1PassDid everything it was expected to do.
0FailNo test case failed.
0Not gradableEvery result here could be graded.
0UntestedEvery test case in this scenario has a result.
The pass rate is a summary. The evidence is the test case below: what we sent, what we checked, what the tool returned, and the proof.

The test case

Each test case is judged on its own: Pass, Fail, Not gradable, or Untested. The scenario result above counts this row.

Clean scan — text recoveredPassEvidence
What we sent
InputDocument (PDF)
Scanned one-page laboratory certificate with visible letterhead, reference and date, six-row results block, signature area, and footer. From outside the graded run; shown for context, not graded.
Open the PDF ↗
What the tool returned
Markdown54 lines
Kirkhollow Analytical Laboratories Ltd

Braithe Road, Vardenne BR7 2HD 01632 960118 · reception@kirkhollow-labs.example

Ms R. Sandiman Braithe Estuary Partnership Estuary House, Vardenne Quay Braithe BR3 4LN

Our reference: KAL-11938/26
Date of report: 4 February 2026

### Certificate of analysis

Dear Ms Sandiman

Thank you for the four water samples delivered to this laboratory on 28 January 2026. They were booked in as batch KAL-11938 and analysed by the methods shown. The results below relate only to the samples as received.

<table>
  <tbody>
    <tr>
      <td>Sampling point</td>
      <td>Vardenne Channel, station VQ1</td>
    </tr>
    <tr>
      <td>Date and time of sampling</td>
      <td>27 January 2026, 09:40</td>
    </tr>
    <tr>
      <td>Total suspended solids</td>
      <td>41.6 mg/l (method KAL-M04)</td>
    </tr>
    <tr>
      <td>Chloride</td>
      <td>18,200 mg/l (method KAL-M11)</td>
    </tr>
    <tr>
      <td>Ammoniacal nitrogen</td>
      <td>0.42 mg/l (method KAL-M12)</td>
    </tr>
    <tr>
      <td>Dissolved organic carbon</td>
      <td>5.9 mg/l (method KAL-M22)</td>
    </tr>
  </tbody>
</table>

The suspended solids figure is higher than those reported for this station in December and January, which were both below 30 mg/l. The samples were taken on the ebb, two days after the last placement at Kellow Sands, and the result is consistent with that.

The remaining samples will be held until 4 March 2026 and destroyed after that date unless you ask us to keep them.

Yours sincerely

Dr A. J. Prosser
Laboratory Manager

Registered in England 2214887 · Results relate only to the samples as received

Copied from Proof 1 · Output file (Markdown), lines 1–54

Open the Markdown file ↗
Proof 1Output file (Markdown)
Markdown output for the certificate_of_analysis_KAL-11938 PDF, showing the letterhead, recipient block, reference and date, six-row results table, closing paragraphs, sign-off, and footer. From outside the graded run; shown for context, not graded.

Its contents are printed above, under What the tool returned.

Open the Markdown file ↗
Expected vs. Found
What this scenario evaluates

These are scenario-level criteria. Each test case's Expected and Found are listed separately.

  • Whether the visible text from the scanned page is recovered accurately.
  • Whether the page is not silently skipped or treated as empty because no text layer exists.
  • Whether the output stays faithful to the visible content without inventing text.
  • Whether scanned pages are handled as real input, not hidden by omission.
How this scenario is judged →
✓Found: All visible text was recovered accurately from the image-only page.
Why this result

The output reproduces the page’s visible text in order, including the letterhead, reference, six results rows, closing text, sign-off, and footer. The handwritten signature is not transcribed.

Also observed on this row

Two-column results table. The results block is returned as a two-column HTML table. Each label appears beside its value.

Tested by Chandresh Bisht · evidence dated 9 September 2026

Configuration and setup

How this tool was set up for the run and what the test needed in place. Each row is a fact from the run's records; a fact the records do not hold is left out, not guessed.

Software that produced the output
Extend AI
Build or version
x-extend-api-version 2026-02-09; engine parse_performance 2.0.0
Plan or tier
free tier (10,000 starter credits, Pay As You Go)
Model
x-extend-api-version 2026-02-09; engine parse_performance 2.0.0
Surface
API
Set up before the run
A short, clearly scanned, image-only PDF with ordinary text was provided; the complete PDF went through the default Markdown conversion pipeline in one pass.
Tested
9 September 2026 · Chandresh Bisht

How this scenario is graded

How we decide Pass, Fail and Not gradable. The same rules apply to every tool tested on this scenario.

How results are decided

Each test case gets one result per tool: Pass, Fail or Not gradable. A test case we haven't run yet shows Untested. There are no partial results.

The rules
  • Pass: the tool did everything the test expected, and nothing it said contradicts the correct answer.
  • Fail: at least one expected behaviour clearly didn't happen; the row says which and quotes the tool.
  • Not gradable: our evidence couldn't settle the outcome (for example a record we needed is missing). It is never counted as a fail, and the row says what's missing.

Other tools on this scenario

How the 13 tools in this benchmark did on this scenario. A tool with a published result links to it.

Open the scenario page →
Adobe PDF Extract API0 of 1 passedNo result page yet
Extend AI1 of 1 passedThis result
Firecrawl0 of 1 passedNo result page yet
GPT-5.6 terra1 of 1 passedNo result page yet
Landing AI1 of 1 passedNo result page yet
LlamaParse1 of 1 passedNo result page yet
Mistral AI →1 of 1 passed
Nutrient.io0 of 1 passedNo result page yet
PDF Vector1 of 1 passedNo result page yet
Reducto →1 of 1 passed
Tensorlake1 not gradableNo result page yet
Unstructured1 of 1 passedNo result page yet
Upstage AI1 of 1 passedNo result page yet

Extend AI on the other scenarios

12 scenarios in this benchmark. A scenario with a published result links to it.

Open the tool page →
A cleanly scanned document1 of 1 passedThis result
A document containing a code block0 of 1 passedNo result page yet
A document containing mathematical equations0 of 1 passedNo result page yet
A document mixing digital and scanned pages1 of 1 passedNo result page yet
A document with a data chart1 of 1 passedNo result page yet
A document with figures and captions1 of 1 passedNo result page yet
A document with styled headings and subheadings0 of 1 passedNo result page yet
An ordinary digital text document1 of 1 passedNo result page yet
A page laid out in multiple columns1 of 1 passedNo result page yet
A simple, clearly formatted table1 of 1 passedNo result page yet
A table that continues across a page break1 of 2 passedNo result page yet
Footnotes at the bottom of the page1 of 1 passedNo result page yet

Where this sits in the benchmark

This page is one cell of a larger study: one tool, one scenario. Only this benchmark's frame appears here.

LevelNameScope
BenchmarkConverting a complex PDF into clean Markdown with a hosted API12 scenarios · 13 tools
CapabilityScanned Document OCR
ScenarioA cleanly scanned document
ToolExtend AI

History of this result

What has happened to this result since it was first published. Runs and grades are never overwritten: a retest or a re-grade publishes a new result and keeps the earlier one readable.

from the publication record
6 October 2026First publishedStructured Document Extraction v1
6 October 2026Updated — retested on x-extend-api-version 2026-02-09; engine parse_performance 2.0.0now under Converting a complex PDF into clean Markdown with a hosted API v1
The result is unchanged: 1 pass · 0 fail. The earlier result stays readable. Read the earlier result →

Act on this result

Nothing filed here edits the run or the grade. A challenge opens a review, and a review can produce a new run or a re-grade — which becomes the current result and leaves this one in the history.

This matches what I see

You run the same kind of test against your own setup and get the same behaviour.

Agree →
This does not match

Yours behaves differently. Tell us what you got, with a screenshot if you have one.

Disagree →
Point out an issue

Something here is wrong — a reference value, a transcription, a grade.

Report an issue →
Request a retest

On a newer build, a larger dataset, or your own setup.

Request a retest →
We have fixed this

Tell us what changed and we schedule a rerun of the failing test case. The old result stays as history.

Vendor notice →
Filed against this evidenceNothing yet. Challenges, counter-evidence and fix notices appear here with their outcome, and stay on the page after they are resolved.
Cite this result
aidemos.com/benchmarks/pdf-to-markdown-apis/results/extend-ai/a-cleanly-scanned-document · 1 pass · 0 fail · coverage 1/1 · graded 2026-09-29

The same record is available as structured data through the AI Demos MCP server, with the counts, the coverage and every per-test-case reason carried as fields.

Verify the proof files

These files support this result. Open a file to inspect the original evidence.

Input · Document (PDF) ↗application/pdf · 747 KB
File fingerprints (SHA-256)

A fingerprint identifies the exact file used for this result.

Proof 1 · Output file (Markdown)5eeacf2c96aa80822f04eba51a32cbaf0ed597ef6711cd855a10b97a88f21390
Input · Document (PDF)a459554d58b9480b19bb3a63598c500b961c3ad377492d18ff227bb439e1f68c