88Z AI, 24 September — China Telecom Xingchen Lab has open-sourced TeleOCR, a document parsing model using a lightweight vision-language architecture with approximately 1.2 billion parameters. It uniformly processes digital and photographed documents, converting text, layout, tables, formulas, and scientific charts into structured results. The model topped the OmniDocBench v1.6 leaderboard with a total score of 96.87, surpassing competitors like MinerU2.5-Pro, and also secured first place on PureDocBench and won the I
At a glance
- TeleOCR uses a 1.2B parameter lightweight architecture to uniformly parse digital and photographed documents into structured results
- The model topped OmniDocBench v1.6 with 96.87 points, and won first place on PureDocBench and the ICDAR-2026 championship
The update
China Telecom Xingchen Lab has officially open-sourced TeleOCR, a document parsing model. Built on a lightweight vision-language architecture with approximately 1.2 billion parameters, it uniformly processes both digital and photographed documents. Its core capability lies in accurately converting text, layout, tables, mathematical formulas, and scientific charts within documents into structured data results. In terms of performance, TeleOCR topped the OmniDocBench v1.6 benchmark with a total score of 96.87, outperforming well-known models such as MinerU2.5-Pro and PaddleOCR-VL-1.6. Furthermore, the model secured first place on the PureDocBench leaderboard and won the championship in the ICDAR-2026 Scientific Chart Parsing Challenge, demonstrating its leading strength in complex document parsing.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 中国电信天翼AISource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.