Nace AI launches document parsing model Drex at $7 per thousand pages
Nace AI has launched the document parsing model Drex. After converting a PDF or scanned document into text, this small model compares each block against the original page image, and only sends the mismatched parts back for correction, at a cost of $7 per thousand pages. Drex takes about 50 milliseconds to answer narrow questions such as "does the OCR result for a given line match the image," at about $2 per million answers; the median time for single-page correction drops by 28% to 38%, and if correction fails, it falls back to the original recognition result. The system scores 0.834 on the Parse Index, higher than GPT-6 Astra (low) at 0.813, at a cost of $7 per thousand pages versus $70.57.
Why read this
AI score 53
The original text provides the cost and benchmark comparison for self-verifying document parsing, which can be used to judge the trade-offs of the small-model error-correction approach in batch scenarios.
Topics
View original postIntroducing Drex for Document Processing 📑 Decision-powered parsing that catches errors before they reach your agents. → #1 on our parsing leaderboard → Beats GPT-6 Astra, Claude Opus 4.7 & every parser tested → 38% faster corrections on RealDocBench Try it: https://t.co/eTUlO9j0wU
View original
Introducing Drex for Document Processing 📑 Decision-powered parsing that catches errors before they reach your agents. → #1 on our parsing leaderboard → Beats GPT-6 Astra, Claude Opus 4.7 & every parser tested → 38% faster corrections on RealDocBench Try it: https://t.co/eTUlO9j0wU
Body
really cool launch.
Brilliant use of Decision models.
@NaceAI built a document reader that proofreads itself. After it turns a PDF or scan into text, a small model called Drex compares each block with the original page image and sends only the mismatches back for correction, all for $7 per 1,000 pages.
Standard OCR flattens tables and drops ticked boxes, while frontier models help but get expensive at volume.
Drex instead answers narrow questions, such as whether an OCR line matches the image, in about 50 milliseconds for about $2 per million answers.
Because only doubted text gets rewritten, median correction time per page fell 28% to 38%, and failed fixes revert to the original reading.
The system scores 0.834 on the Parse Index, beating GPT-6 Astra (low) at 0.813 while costing $7 per 1,000 pages against $70.57.
https://t.co/wnYo5UWUwz
AI Trends