krutrim-ai-labs/Chitrapathak-2

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:Mar 11, 2026License:otherArchitecture:Transformer0.0K Featherless Exclusive Cold

Chitrapathak-2 by krutrim-ai-labs is a 3 billion parameter Vision-Language Model (VLM) based multilingual Optical Character Recognition (OCR) system, specifically designed for the linguistic diversity of the Indian ecosystem. It excels at high-fidelity OCR for 10 major Indian languages and English, particularly on printed book pages and PDF documents. This model offers strong generalization across diverse font styles and layouts, demonstrating state-of-the-art OCR accuracy in Telugu and competitive performance across other Indic languages.

Loading preview...

Chitrapathak-2: Multilingual OCR for Indic Languages

Chitrapathak-2 is a 3 billion parameter Vision-Language Model (VLM) developed by krutrim-ai-labs, specifically engineered for high-fidelity Optical Character Recognition (OCR) across the diverse linguistic landscape of India. It is the second iteration in the Chitrapathak series, focusing on robust OCR for Indic scripts and multilingual documents.

Key Capabilities

  • Multilingual Support: Processes 10 major Indian languages (Hindi, Sanskrit, Bengali, Telugu, Tamil, Marathi, Kannada, Malayalam, Odia, Punjabi) and English.
  • Optimized for Document Types: Excels at OCR for printed text in multilingual books, PDF documents, and scanned text.
  • Layout Robustness: Demonstrates strong performance on complex layouts including multi-column, multi-font, and dense paragraphs.
  • High Accuracy: Achieves state-of-the-art OCR accuracy in Telugu and is highly competitive with top models like Gemini-2.5 across other Indic languages on the IndicVisionBench-OCR benchmark.
  • Integration: Compatible with Hugging Face Transformers and vLLM for flexible deployment.

Good For

  • Digitizing Historical Texts: Ideal for converting scanned book pages and archival documents in Indic languages into editable text.
  • Multilingual Document Processing: Suitable for applications requiring OCR of documents containing a mix of English and various Indian languages.
  • Research and Development: Provides a strong foundation for further research in multilingual OCR, especially for under-resourced languages.

Limitations

  • Primarily designed for OCR; not intended for broader document intelligence tasks.
  • Performance may degrade on handwritten, noisy, or low-resolution images.
  • Less effective on rare Indic/English scripts or non-book domains (e.g., forms) and complex index-page layouts.

Finetunes
1 models