ProKope-4B-Eagle-Eye — Document Parsing & Table Reconstruction

4.2B Vision-Language Engine (`Qwen3_5ForConditionalGeneration`, `bfloat16`, 32K Context) · Single-Pass PDF & Image to Markdown, HTML Tables & LaTeX
#1 Global · llamaindex/ParseBench (86.01%) #1 Global · ParseBench Tables (93.95%) & Layout (90.46%) #3 Global / #1 4B · allenai/olmOCR-bench (87.06%) #1 Global · ArXiv Math (95.56%) & Old Scans (85.19%)
Model Card & Access Request Commercial & Enterprise Inquiries 
{}

All benchmark scores are indexed on Hugging Face's official evaluation leaderboards: 


        

      

    


    

      

        
1. Self-Hosted / On-Premise Commercial Node License

        

          Deploy ProKope-4B-Eagle-Eye inside your own private VPC, AWS/GCP cluster, or air-gapped on-premise server (HIPAA, SOC 2, LegalTech, FinTech & Defense). Runs on a single 16GB–24GB GPU via vLLM, transformers, or our turnkey Docker appliance.
        

        Submit Corporate Access Request
      


      

        
2. Managed Enterprise Parse API & Custom Evaluation Trials

        

          Integrate high-throughput multi-page PDF and image extraction (POST /v1/parse) into your production RAG or document automation pipeline without managing GPU infrastructure. Dedicated trial API keys available for qualified engineering teams.
        

        Contact prokope.model@gmail.com