PaddlePaddle Introduces Unlimited-OCR, a New OCR Model Designed for Large Document Processing.
PaddlePaddle Introduces Unlimited-OCR, a New OCR Model Designed for Large Document Processing.
PaddlePaddle has announced the Unlimited-OCR model, capable of processing large documents without speed loss.
According to the developers, the model can process hundreds of pages in a single pass without noticeable speed loss. This has been made possible by the R-SWA (Reference Sliding Window Attention) mechanism, which maintains a constant size of the KV cache during decoding.
In the OmniDocBench benchmark, the model scored 93%, outperforming DeepSeek-OCR by 6%.
Why it matters
AnalysisThis new model can significantly simplify the processing of large documents, marking an important step in the development of OCR technologies. Its high performance and accuracy open up new possibilities for document processing automation.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
LLM (Large Language Model)
A fundamental class of neural network architectures based on autoregressive transformers, predicting the probabilistic distribution of subsequent tokens and demonstrating emergent properties of abstract reasoning, code synthesis, and logical inference.
Jev and System One AI: A Complete Practical Guide to the Decision Model from TypeSafe
A deep engineering breakdown of the flagship Jev model: System One architecture versus LLMs, Choice/Score/Noul primitives, request parallelism, confidence mathematics, video use cases, code, and an overview of local open-weight alternatives (Laya, GLiNER2.5, CLM-8B).
Part II. AI Agent Architecture: From Model to Memory and Tools
A deep engineering dive: tokenization, KV-cache, sampling determinism, the training stack, quantization, context engineering, the safety stack, production RAG, ReAct, and multi-agent swarm systems.
Jev Decision Model
Jev is a commercial non-autoregressive System 1 model from TypeSafe AI for fast typed decisions in agentic loops: it returns a validated decision object in 70–500 ms at ~$0.042/1M input tokens, eliminating JSON syntax hallucinations and high-latency routing in ReAct loop/Tool Calling stacks.