prithivMLmods/visionOCR-3B-061125

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:Jun 11, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The prithivMLmods/visionOCR-3B-061125 model is a 3 billion parameter vision-language model, fine-tuned from Qwen/Qwen2.5-VL-3B-Instruct. It specializes in Document-Level Optical Character Recognition (OCR), long-context vision-language understanding, and accurate image-to-text conversion, including mathematical LaTeX formatting. This model excels at extracting structured content from complex documents and performing visual reasoning across diverse input formats, making it ideal for document comprehension and structured data extraction tasks.

Loading preview...

Overview

The prithivMLmods/visionOCR-3B-061125 is a 3 billion parameter vision-language model, fine-tuned from the Qwen2.5-VL architecture. It is specifically optimized for advanced Document-Level Optical Character Recognition (OCR), long-context vision-language understanding, and precise image-to-text conversion, including mathematical LaTeX formatting. This model significantly enhances document comprehension, structured data extraction, and visual reasoning across various input formats, offering competitive performance on benchmarks like DocVQA, MathVista, RealWorldQA, and MTVQA.

Key Capabilities

  • Advanced Document-Level OCR: Extracts structured content from complex, multi-page documents such as invoices, academic papers, and scanned reports.
  • Enhanced Long-Context Vision-Language Understanding: Handles dense document layouts, long sequences of embedded text, tables, and diagrams with coherent cross-reference understanding.
  • Mathematical LaTeX Formatting: Accurately extracts and formats mathematical expressions from printed or handwritten content.
  • Video Understanding: Supports detailed comprehension of long-duration videos (up to 20+ minutes) for content summarization and Q&A.
  • Visually-Grounded Device Interaction: Enables mobile/robotic device operation via visual inputs and text-based instructions.

Good for

  • High-fidelity OCR from diverse documents, forms, receipts, and scanned materials.
  • Image and document-based question answering for educational and enterprise applications.
  • Retrieval and summarization from long documents, slides, and multi-modal inputs.
  • Multilingual OCR and structured content extraction for global use cases.
  • Robotic or mobile automation requiring vision-guided contextual interaction.

Finetunes
2 models