Enterprise DNAEnterprise DNA
← Back to the Model Guide

PRACTICAL WORKFLOW / 04

Extract document data.

Extract structured fields from scans, invoices and reports, with evidence for each value.

Vision support does not guarantee PDF upload support. Check your app, and review low-quality scans manually.

Three ways to choose.

Explore matching models ↗

These are automated capability-based shortlists, not Enterprise DNA tested recommendations. Candidates require text support and at least 32,000 context tokens. Preview, experimental, batch and free variants are excluded. Cost picks require non-zero listed rates. Published index scores are a guide, not proof of task completion.

Published benchmark leader

No matching measurement is available. Explore the catalogue and test a candidate on your own sample.

Lowest estimated API cost

S

Extract ↗

Lowest listed cost for 10,000 input and 2,000 output tokens. Output quality and retries can change the final cost.

Input / 1M tokens
$0.040
Output / 1M tokens
$0
Example task estimate
$0.00040 USD
Price record
2026-10-18
Check source & availability ↗

API estimates exclude caching, tools, reasoning overhead, retries and provider fees. Access through a chat app or subscription has its own pricing and capabilities.

GET A USEFUL RESULT

Your first run.

  1. Prepare a redacted sample with a known correct answer.
  2. Specify the fields and require page references.
  3. Validate the structure and compare every extracted value.

A prompt to start with.

Extract the following fields from the attached document into JSON: [field names]. For each field include value, page and the exact supporting text. Use null when missing or unreadable. Do not infer missing values. Return a separate list of ambiguities for human review.

MAKE YOUR OWN COMPARISON

Same task. Same evidence. Better decision.

Try the same redacted sample and prompt with two models. Record whether each result is correct, how much editing it needs, elapsed time and total cost including retries. Keep the outputs alongside your expected answer. Choose the model that completes your work reliably.

Compare models for this job →