OXR makes a strong open-source debut! 1.3B sweeps to first place on three global leaderboards, faster than Paddle and MinerU, RSI pushes it to new heights

What happened: SoMark open-sourced OXR, a 1.3B parameter document parsing model that ranks first on OmniDocBench, olmOCR-Bench and DocParsingBench. It parses 13 document element types and outputs Markdown or JSON with layout coordinates. Key numbers: 1.3B parameters. Scores of 97.18, 87.84 and 87.76 on the three benchmarks. Why it matters: The team reports top accuracy and faster full parsing than PaddleOCR-VL-1.6 and MinerU2.5-Pro, positioning a lightweight model as a strong open-source option for RAG and Agent document workflows.

OXR's 1.3B model debuts open-source, topping three global leaderboards faster than Paddle and MinerU.

OXR's 1.3B model debuts open-source, topping three global leaderboards faster than Paddle and MinerU. — impact context not separately stored; see What changed.