AI BriefingPick
Unlimited-OCR, a 3.3B-parameter multilingual OCR model, released
·2026.06.24 14:53
Key point
A 3.3B-parameter open-source OCR model capable of parsing entire documents from single images to PDFs has been released.
Details
ModelScope has released Unlimited-OCR. This model adopts a full-document parsing approach rather than the conventional region-based cropping method, enabling more accurate understanding of document structure.
Key features include:
- 32K output length: Supports processing long OCR sequences.
- Support for diverse layouts: Provides optimized performance for different document layouts through Base and Gundam image modes.
- Efficient serving: Supports Transformers inference and SGLang serving, with OpenAI-compatible streaming requests available.
- Open source: Provided under the MIT license, allowing free commercial use and modification.
This model is designed to advance document parsing technology in the style of DeepSeek-OCR to the next level.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
GitHub Next releases Ace, a local collaborative coding environment for humans and AI agents