HunyuanOCR 1.5
HunyuanOCR 1.5 is Tencent's update to its lightweight end-to-end OCR model, keeping the 1.0 backbone and making it faster and better on long-tail tasks.
A release of Hunyuan OCR.
- Released
- July 7, 2026
- Model ID
- tencent/HunyuanOCR
- Size
- Same backbone as HunyuanOCR 1.0, plus a small DFlash draft model
- License
- Tencent Hunyuan Community License (same as 1.0)
- Also searched as
- hunyuanocr 1.5, hunyuan ocr 1.5, HunyuanOCR-1.5
What is new in HunyuanOCR 1.5
- DFlash speculative decoding: a small block-diffusion draft model proposes tokens that the main model checks, keeping the same output distribution.
- 6.37x faster inference with Transformers and 2.14x with vLLM, per the paper.
- Runs on a PC through llama.cpp and GGUF.
- Maximum image resolution raised to 4K and context window to 128K.
- "Agentic Data Flow" training data for ancient scripts, low-resource languages, charts, tables and multi-image questions.
HunyuanOCR 1.5 benchmarks
| Benchmark | Score | Note |
|---|---|---|
| Inference speedup, Transformers | 6.37x | |
| Inference speedup, vLLM | 2.14x |
Scores as published by the authors. We have not rerun them.
Use HunyuanOCR 1.5 through an API
OCRRouter runs a different release of Hunyuan OCR as tencent/hunyuan-ocr.
1DeepSeek-OCROpen-source vision model that compresses pages into few tokens.1 credit / page
deepseek/deepseek-ocr1 credit / page2PaddleOCR-VLOpen-source document model from Baidu. Excellent on Chinese, Japanese and Korean.6 credits / pagepaddle/paddleocr-vl6 credits / page3GLM-OCRSmall open-source OCR model from Z.ai with layout regions. MIT licensed.6 credits / pagezai/glm-ocr6 credits / pageHunyuan OCR versions
- Hunyuan OCR: all versions, price, speed and features on OCRRouter
- HunyuanOCR 1.5 (July 7, 2026), this page
Questions
Sources
Last updated .