Baidu has open-sourced ERNIE OCR 2, a new-generation document-parsing model with 3 billion total parameters and only 0.5 billion activated, which tops authoritative benchmarks on hundred-page-long documents, an October 29 AI-daily roundup reported.
The model is designed for high-accuracy extraction from long, complex documents, addressing a persistent bottleneck in turning scanned and layered files into structured, machine-readable data for enterprise and agent workflows.
The same report said Baidu is set to disclose progress on world-model research and development, with the capability expected to be applied first in autonomous-driving scenarios, where predicting and simulating physical environments is central.
The moves continue a streak of Chinese frontier-model open-sourcing in which leading labs release capable weights to seed ecosystems and accelerate downstream application, rather than keeping them behind APIs.
Source: AI-daily roundup / Baidu (paraphrased and translated from Chinese technology-media reporting; reprinted for informational purposes with attribution).
