Unlimited-OCR reads 100-page PDFs in one shot
Shruti_0810 · x · 2026-07-20
China has open-sourced Unlimited-OCR, a 3B model for reading full documents in one pass.
- Uses a 32K context window to process an entire document at once instead of page by page
- Preserves cross-page context, which helps with tables, references, and long-form documents
- Reports 93% on standard OCR benchmarks, about +6 over the baseline
- Claims <0.11 error rate beyond 40 pages
- Runs fully locally and supports Transformers, vLLM, Ollama, Docker, llama.cpp, and more
The post argues this approach makes enterprise OCR more reliable while eliminating per-page cloud OCR costs.
Related event: Baidu Open-Sources Unlimited-OCR for Long Documents(4 posts)→
More from Multimodal
- 10 practical ways to run Astra computer use on Premiere Pro for video editing — beechinour · 2026-10-27
- Krea ships Premiere Pro plugin to generate video and SFX inside the timeline — lmoroney · 2026-10-27
- Wan2GP lands on Pinokio: 1-click AI video on 6GB GPUs — cocktailpeanut · 2026-10-27
- WanGP v18.81 adds Tencent's Hy-Nomas world model, playable on 8GB VRAM — cocktailpeanut · 2026-10-27
- Butter.video wins over skeptical creator with hundreds of tasteful timeline 'blocks' for video effects — beechinour · 2026-10-27
- AI Video Pits Brock Lesnar Against Duncan the Tall in a Toy Showdown — KloneStudios · 2026-10-27