HunyuanOCR-1.5
The current tencent/HunyuanOCR repository documents HunyuanOCR-1.5. It retains the earlier lightweight backbone and upgrades document training and inference.
Model design and access
DFlash uses a block-diffusion draft model followed by target-model verification. The training update targets low-resource scripts, historical text and multi-image text questions. The project supports server inference and GGUF-based CPU or consumer-GPU deployment.HunyuanOCR-1.5 official model documentation
Practical use
Practical guidance. Pin a repository revision because the repository name does not contain the 1.5 version. Compare complete structured outputs with and without speculative decoding, and measure the benefit on long tables rather than only short text.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| License | Tencent Hunyuan Community License; custom terms applyHunyuanOCR-1.5 official model documentation |
| Context | 128K tokensHunyuanOCR-1.5 official model documentation |
| Image resolution | Up to 4KHunyuanOCR-1.5 official model documentation |
Related models
- Youtu-Parsing-Omni
Tencent ยท Youtu-Parsing
Youtu-Parsing-Omni is a 5B model that returns structured perception and description data from visual or audio input.
Launch date not confirmedDownloadable