SoMark’s 1.3B OXR tops three document parsing benchmarks with recursive self-improvement

What happened

SoMark has open-sourced OXR, a 1.3B document parsing model that reportedly scores 97.18 on OmniDocBench, 87.84 on olmOCR-Bench, and 87.76 on DocParsingBench, ranking first on all three. Its recursive self-improvement process identifies errors and uncertain predictions, clusters their causes, and generates targeted training data without new manual annotations. OXR separates layout analysis from recognition and reading order prediction, reportedly parsing 100 pages in 101.0 seconds on one RTX PRO 6000; code, weights, evaluation procedures, and a free API requiring no registration or key are available.

Read next

• More changes in today’s briefingMistral releases improved Mistral Large 4 open weights for coding and cybersecurity

The story so far

SoMark has open-sourced OXR, a document parsing system with a 1.3B-parameter main model, making its code, weights and benchmark evaluation procedures publicly available. It also offers an online demo and a free API requiring no registration or API key. SoMark said it would maintain the free service while server load remains manageable. OXR handles 13 document element types and outputs Markdown or JSON with layout coordinates. It separates replaceable layout analysis from recognition and reading order prediction. Its recursive self-improvement process identifies errors and uncertain predictions, groups their causes and generates targeted training data without new manual annotations. SoMark reported first-place scores of 97.18 on OmniDocBench, 87.84 on olmOCR-Bench and 87.76 on DocParsingBench. The team also said OXR completed full document parsing faster than PaddleOCR-VL-1.6 and MinerU2.5-Pro under the same test conditions, and reported parsing 100 pages in 101.0 seconds on a single RTX PRO 6000.