I also get connection timeouts on larger documents, but it automatically retries and completes. All the pages are processed when I'm done. However, I'm using the Python client SDK for larger documents rather than the basic glmocr command line tool. I'm not sure if that makes a difference.