OvisOCR2 Technical Report

2026-07-15

AI Summary

This work proposes an end-to-end document image parsing method that directly converts page images containing text, mathematical expressions, tables, and visual elements into Markdown following natural reading order. By integrating real annotated data with HTML-generated synthetic data, the approach trains a unified 0.8B-parameter model through supervised fine-tuning, multi-component reward-based reinforcement learning leveraging a 4B-scale model, online policy distillation, and model fusion. The resulting system achieves state-of-the-art performance on OmniDocBench v1.6 with a total score of 96.58—the first end-to-end architecture to top this benchmark—and attains the highest Avg3 score (75.06) on PureDocBench, significantly outperforming conventional multi-stage pipelines and demonstrating exceptional generalization on long-tail, challenging document scenarios.

Abstract

We introduce OvisOCR2, a 0.8B document parsing model. OvisOCR2 is designed as an end-to-end parser: given a document page image, it generates a Markdown representation in natural reading order, covering text, formulas, tables, and visual regions. We build a data engine that combines filtered real-document annotations with synthetic pages whose rendered images and Markdown targets are derived from the same HTML source. The training recipe includes supervised fine-tuning, reinforcement learning on a 4B branch with a multi-component reward design, on-policy distillation into the 0.8B model, and model fusion. On OmniDocBench v1.6, OvisOCR2 achieves a state-of-the-art overall score of 96.58, placing an end-to-end model at the top of this leaderboard previously dominated by pipeline methods and highlighting the potential of end-to-end document parsing. On PureDocBench, OvisOCR2 also achieves the highest Avg3 score of 75.06. Beyond these two public benchmarks, we evaluate OvisOCR2 on an in-house benchmark designed to cover a broader set of long-tail and challenging scenarios. OvisOCR2 obtains the best overall performance among the compared methods, providing further evidence of its generalization and robustness. OvisOCR2 is available at https://huggingface.co/ATH-MaaS/OvisOCR2.

Problem

• document parsing
• end-to-end
• OCR
• Markdown generation
• layout understanding

Innovation

• end-to-end document parsing
• synthetic data engine
• reinforcement learning with multi-component reward
• on-policy distillation
• Markdown generation