China Telecom's Xingchen General Artificial Intelligence Laboratory has unveiled TeleOCR, a lightweight model for document parsing that boasts approximately 1.2 billion parameters. This model has excelled in numerous public evaluations and the relevant ICDAR 2026 competitions, and is now accessible as open-source software, complete with online demonstrations. TeleOCR effectively tackles real-world challenges in document parsing, such as document deformation, intricate layouts, tables, formulas, and scientific charts. Leveraging technologies like deformation-aware learning, an automated multi-node consistency voting data engine, and content-structure decoupled learning, all underpinned by a four-stage training system, TeleOCR achieves high-precision parsing with a relatively small parameter set. It can transform diverse documents into structured formats, thereby facilitating business processes such as knowledge base construction and RAG (Retrieval-Augmented Generation) applications. Its lightweight design aligns with the need for cost-effective deployment, serving as a pivotal element of China Telecom's foundational AI capabilities in layout analysis, in line with the 'Cloud Transformation, Digital Transformation, Intelligent Empowerment' strategy. TeleOCR reduces the barriers to technological adoption and offers robust support for intelligent document parsing across various industries.