In the specialized field of Optical Character Recognition (OCR), a new open-source model from Datalab is setting a new benchmark for accuracy and versatility. Chandra OCR, released in October 2025, has rapidly ascended to the top of the leaderboards, outperforming even proprietary giants like GPT-4o and Gemini Pro on key benchmarks.
Beyond Simple Text Extraction
Chandra is not just another OCR tool; it’s a comprehensive document AI solution. Unlike traditional pipeline-based approaches that process documents in chunks, Chandra utilizes full-page decoding. This allows it to understand the entire context of a page, leading to significant improvements in accuracy and layout awareness.
Key Capabilities:
- Layout-Aware Output: Chandra preserves the original document structure, outputting to Markdown, HTML, or JSON with remarkable fidelity.
- Image & Figure Extraction: It can identify, caption, and extract images and figures from within a document.
- Advanced Language Support: Chandra supports over 40 languages and can even read handwritten text, making it a truly global solution.
- Specialized Content: The model excels at handling complex content, including mathematical equations and intricate tables.
Unrivaled Performance
| Category | Score | Rank |
|---|---|---|
| Tables | 88.0 | #1 |
| Old Scans Math | 80.3 | #1 |
| Old Scans | 50.4 | #1 |
| Long Tiny Text | 92.3 | #1 |
| Base Documents | 99.9 | Near-Perfect |
Chandra’s performance on the independent olmOCR benchmark is nothing short of revolutionary. With an overall score of 83.1%, it has established a new state-of-the-art for open-source OCR models.
Source: https://medium.com/data-science-in-your-pocket/chandra-ocr-beats-deepseek-ocr-47267b6f4895
Accessible and Production-Ready
Datalab has made Chandra widely accessible. It is available as an open-source project on GitHub and Hugging Face, and also as a hosted API with a free tier for developers to get started. For high-throughput applications, quantized versions of the model are available for on-premises deployment, capable of processing up to 4 pages per second on an H100 GPU.
Why Chandra OCR Matters
The release of Chandra OCR is a watershed moment for document AI. It provides a free, open-source, and commercially viable alternative to expensive proprietary solutions, without compromising on performance. For developers and businesses that rely on accurate and structured data extraction, Chandra OCR is a game-changer.
Read more
Cross-posted from https://www.linkedin.com/pulse/chandra-ocr-best-open-source-ai-document-parsing-matthew-aberham-3fx1e
