Model index

Unlimited OCR 2

Baidu ERNIE’s long-document parsing model, taking page images and returning structured text with cross-page table and paragraph continuity; Baidu describes its approach as unlimited-horizon parsing.

PreviewMultimodalOpen weights

Baidu’s official `baidu/Unlimited-OCR-2` weights were released on 2026-10-27 under MIT; Baidu’s hosted document-parsing service entered public beta at the same time.

MODEL NUMBER043
First public date
2026-10-27
Organization
Baidu
Family
ERNIE / 文心
01

Specifications

Only published specifications are shown.

Input modalities
Text · Image↗
Output modalities
Text↗
Open weights
Yes↗
License
MIT↗
Status
Preview↗
02

Overview

Release dates

— Baidu released ERNIE Unlimited OCR 2 with MIT-licensed weights on 2026-10-27; the hosted document-parsing service is separately in public beta. ↗

Why it mattered

The second model in the Unlimited OCR line, extending image-text understanding toward continuous parsing of entire long documents. Its weights are available under MIT while the hosted parser remains in public beta; vendor benchmark rankings are not treated as independent conclusions.

03

Architecture & capabilities

Model capabilities

VisionData extractionComplex layouts
04

Release history

Open weights

Baidu released ERNIE Unlimited OCR 2 with MIT-licensed weights on 2026-10-27; the hosted document-parsing service is separately in public beta.

View source
05

Related models

Select a card to view the related model.

No related models recorded yet

06

Sources

Suggest a correction

Copy this template, add the field, proposed value, and primary source, and share it with the archive maintainer.