88Z AI, 24 July — Alibaba Cloud has released the open-source document parsing model OvisOCR2, which set a new record of 96.58 on the OmniDocBench v1.6 leaderboard, becoming the first end-to-end model to surpass traditional pipeline methods and take the top spot.
At a glance
- OvisOCR2 set a new record of 96.58 on OmniDocBench v1.6, becoming the first end-to-end model to top the leaderboard.
- Built on 0.8B parameters and hybrid data training, it enables single-step Markdown output for text, formulas, and tables.
The update
Alibaba Cloud has officially open-sourced the lightweight document parsing model OvisOCR2. With only 0.8 billion parameters, the model was developed through post-training based on Qwen3.5-0.8B. In terms of performance, OvisOCR2 achieved a high score of 96.58, successfully setting a new record on the OmniDocBench v1.6 leaderboard. It is the first end-to-end model to surpass traditional pipeline methods and claim the top position. By employing an engine that combines real and synthetic data along with a multi-stage training process, the model efficiently handles complex document content. It supports the direct conversion of various elements, including text, mathematical formulas, and tables, into single-step Markdown output, significantly improving the efficiency and accuracy of document parsing.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 阿里云Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.