Baidu Releases Unlimited-OCR, an MIT-Licensed 3B Model for Document Parsing
Baidu released Unlimited-OCR on June 22, 2026. The 3B vision-language model builds on DeepSeek-OCR and parses images, multipage documents, and PDFs in one pass up to 32,768 tokens. It works with vLLM and SGLang and scored 86.81 on LlamaIndex ParseBench for text content. It reached roughly one million Hugging Face downloads shortly after release.
Visit original source