Unlimited-OCR

Baidu's one-shot long-horizon document parsing model, positioned to "push DeepSeek-OCR one step further": a compact DeepSeek-OCR-derived MoE decoder (12 layers, 64 routed + 2 shared experts, 6 active per token, 32K context) behind a 2048-dim vision encoder. MIT-licensed; total parameter count not stated on the card (~3B-class). The breakout OCR release of June 2026 — 885K HuggingFace downloads and 1,600+ likes within two weeks. Distinct lineage from Qianfan-OCR (a 4B "Layout-as-Thought" VLM from March 2026).