Infinity-Parser2 Technical Report

2026-07-08

AI Summary

This work addresses the scarcity of high-quality annotated data in end-to-end document parsing by proposing a joint optimization framework that integrates controllable data synthesis with multi-task reinforcement learning. The authors develop a scalable synthetic data engine and release a bilingual document dataset comprising 5 million samples. They introduce a verifiable multi-task reward mechanism that unifies perception, structural understanding, and reasoning capabilities, and present two model variants balancing low latency and high accuracy. Evaluated on olmOCR-Bench and ParseBench, the proposed method achieves scores of 87.6% and 74.3%, respectively, significantly outperforming state-of-the-art models such as DeepSeek-OCR-2, and demonstrates strong generalization across tasks including table recognition, chemical formula parsing, and document visual question answering.

Abstract

We present Infinity-Parser2, a large multimodal model that couples a controllable data-synthesis pipeline with multi-task reinforcement learning for end-to-end document parsing, addressing the persistent scarcity of faithfully annotated parsing corpora. Our contributions are threefold. First, we build a scalable synthesis engine, pairing a controllable rendering framework with an iterative refinement loop, and use it to construct and open-source Infinity-Doc2-5M: a 5-million-sample bilingual (Chinese/English) corpus spanning diverse document types, annotated with element bounding boxes, canonical content forms (Markdown, HTML, LaTeX, SMILES, structured charts), and full-page reading order. Second, we introduce a verifiable, multi-task reward system that enables Joint Reinforcement Learning across eight co-trained objectives (document parsing, layout analysis, table parsing, math formula parsing, chart parsing, chemical formula parsing, document VQA, and general multimodal understanding), unifying perception, structure, and reasoning in a single optimization signal. Third, we release two variants under a shared architecture: Infinity-Parser2-Flash, optimized for low-latency inference with a $3.68\times$ throughput gain over Infinity-Parser-7B, and Infinity-Parser2-Pro, engineered for precision-critical settings. Infinity-Parser2-Pro reaches state-of-the-art 87.6% on olmOCR-Bench and 74.3% on ParseBench, surpassing DeepSeek-OCR-2, PaddleOCR-VL-1.5, and MinerU2.5, with strong generalization to charts, chemical formulas, and document VQA.

Problem

• document parsing
• annotated corpora
• data scarcity
• multimodal understanding
• structured document analysis

Innovation

• controllable data synthesis
• multi-task reinforcement learning
• document parsing
• multimodal foundation model
• structured document understanding