OXR leads in document parsing benchmarks with RSI-powered model

OXR: OXR is an open-source document parsing model that converts documents into structured, agent-friendly data. It is the first such parser to incorporate RSI during post-training, where a vision-language model evaluates outputs, rollout disagreements highlight uncertainty, and a clustering model groups issues to guide targeted data generation in subsequent rounds. The development is directly tied to the news announcement highlighting its leading position among open-source options on key document parsing benchmarks and its efficiency advantages. Somarkai: Somarkai is the X account that announced and promoted OXR, quoting details about its training innovations and benchmark leadership. The account is directly associated with the project's release and encourages community engagement through GitHub starring. Training Process: A vision-language model assesses outputs while a separate clustering model identifies uncertainty causes to inform new data generation rounds. Data Output Focus: The model is designed to transform documents into structured formats suitable for AI agents. Parsing Innovation: OXR is the first document parser to integrate RSI into its post-training workflow.