It would be nice to see how it performs on this benchmark: https://github.com/opendatalab/OmniDocBench