# Tencent Hunyuan Open-Sources End-to-End OCR Large Model HyOCR-1.5

- Category: Flash (快讯)
- Published: 2026-07-13T11:56:00.000Z
- Source date: 2026-07-13
- HTML: https://www.88z.ai/en/news/flash-20260713-6eaa312cf9e75b6f
- Markdown: https://www.88z.ai/en/news/flash-20260713-6eaa312cf9e75b6f.md
- Other language: https://www.88z.ai/news/flash-20260713-6eaa312cf9e75b6f.md

Tencent Hunyuan has released HyOCR-1.5, the industry's first fully open-source end-to-end OCR expert large model for both training and inference. By introducing DFlash speculative decoding, the model significantly accelerates inference speeds and ranks first in end-to-end tasks on the OmniDocBench v1.6 benchmark with a score of 94.74. HyOCR-1.5 supports eight types of tasks, including document parsing, ancient character recognition, and multi-page Q&A, covering 331 languages.

## At a glance

- HyOCR-1.5 is the industry's first fully open-source end-to-end OCR large model, ranking first on OmniDocBench v1.6.
- It introduces DFlash speculative decoding, boosting inference speed by 6.37x on Transformers, and supports 331 languages across 8 task types.

## The update

Tencent Hunyuan has officially open-sourced HyOCR-1.5, the industry's first end-to-end OCR expert large model with full openness for both training and inference. In terms of performance optimization, the model introduces DFlash speculative decoding technology, which increases inference speed by 6.37 times under the Transformers framework and by 2.14 times in the vLLM environment. With its superior performance, HyOCR-1.5 achieved a score of 94.74 on the OmniDocBench v1.6 benchmark, ranking first in end-to-end tasks. Regarding functional applications, HyOCR-1.5 demonstrates broad applicability, capable of handling eight major categories of tasks such as document parsing, ancient character recognition, chart parsing, and multi-page Q&A. Furthermore, the model supports up to 331 languages, meeting diverse global multilingual document processing needs and providing developers with a powerful open-source OCR solution.

## Sources

- [腾讯混元](https://mp.weixin.qq.com/s/vKFCa9FfoGBUGK8J1MhFag) (2026-07-13)

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.