SoMark, a startup focused on document intelligence, has open-sourced its document parsing model OXR, according to Techub News. The model has 1.3 billion parameters and ranked first on three benchmark tests: OmniDocBench, olmOCR-Bench, and DocParsingBench. SoMark said OXR can parse 13 types of document elements, including tables, formulas, and chemical expressions, and can output results in Markdown or JSON formats.
In a speed test, OXR processed a 100-page document in about 101 seconds on average, outperforming PaddleOCR-VL-1.6 and MinerU2.5-Pro in the same test. The team said the model separates layout analysis from content recognition in its architecture and introduces a recursive self-improvement training paradigm for document parsing during post-training to improve performance on complex documents.
SoMark has released the code and model weights for OXR and is also offering an online demo and a free API. The team said the goal is to turn knowledge embedded in documents into high-quality data that AI systems can use in more real-world business scenarios. The item was cited by QbitAI.
SoMark, a startup working on document intelligence, has open-sourced its document parsing model OXR, according to Techub News.
Top ranking across three benchmarks
OXR has 1.3 billion parameters and ranked first on OmniDocBench, olmOCR-Bench, and DocParsingBench. The model can parse 13 types of document elements, including tables, formulas, and chemical expressions, and export results in Markdown or JSON.
About 101 seconds for 100 pages
In parsing speed tests, OXR took about 101 seconds on average to process a 100-page document, faster than PaddleOCR-VL-1.6 and MinerU2.5-Pro in the same test.
Architecture splits layout and recognition
SoMark said OXR separates layout analysis from content recognition in its architecture. The team also introduced a recursive self-improvement training paradigm for document parsing during post-training for the first time, with the stated aim of improving performance on complex documents.
Code, weights, demo and free API released
SoMark has made the code and model weights available and is also offering an online demo and a free API. The team said the move is intended to make knowledge inside documents into high-quality data usable by AI and bring it into more real business scenarios. The report cited QbitAI.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.