如题，求问靠谱吗？这个ocr能力也太差了，远远不及qwen2.5 72b。具体体现在对于影印的pdf直接识别不出来，输出“It seems you forgot to include the PDF page content. Please provide the text or content of the PDF page you’d like me to convert to markdown. I’ll get back to you with the markdown.” 
完全不是https://getomni.ai/blog/benchmarking-open-source-models-for-ocr中显示的排名 
 [image] 
我用https://github.com/getomni-ai/zerox/tree/main直接调的api