PulseAugur
EN
LIVE 08:45:35

Baidu's Unlimited OCR processes 40+ pages with novel memory mechanism

Baidu has developed an "Unlimited OCR" system capable of processing over 40 pages of documents in a single pass, a significant improvement over previous systems that could handle around ten. This advancement is achieved through a modified attention mechanism that maintains a constant memory usage, regardless of the document length. The system also boasts a processing speed of 5,580 tokens per second and currently leads the most important OCR benchmark. AI

IMPACT This advancement in OCR technology could significantly speed up document processing and data extraction for businesses and researchers.

RANK_REASON New model release from a frontier lab (Baidu) with a novel technical approach and benchmark claims. [lever_c_demoted from frontier_release: ic=2 ai=1.0]

Read on The Decoder →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

Baidu's Unlimited OCR processes 40+ pages with novel memory mechanism

COVERAGE [2]

  1. The Decoder TIER_1 English(EN) · Jonathan Kemper ·

    Baidu's "Unlimited OCR" processes dozens of document pages in one pass by treating memory like human forgetting

    <p><img alt="Abstract collage featuring a wavy strip of text running through a glowing triangle, symbolizing OCR-based text recognition." class="attachment-full size-full wp-post-image" height="1047" src="https://the-decoder.com/wp-content/uploads/2026/07/baidu-ocr-machine-nano-b…

  2. Mastodon — mastodon.social TIER_1 Polski(PL) · aisight ·

    Baidu engineers developed Unlimited OCR – a model that processes over 40 pages of documents with controlled forgetting

    Inżynierowie z Baidu opracowali Unlimited OCR – model, który dzięki mechanizmowi „kontrolowanego zapominania” przetwarza ponad 40 stron dokumentów przy zachowaniu stałej prędkości 5 580 tokenów na sekundę. # si # ai # sztucznainteligencja # wiadomości # informacje # technologia h…