TeleOCR (ex NaviDC-OCR)
Lightweight open document-parsing VLM. Technical report: arXiv 2608.12898.
PricePrice not published
Checked6 October 2026
No published training cutoff
At least1,834events since its release on 17 August 2026
78 historic, 424 major. 97 days of news.
TeleAI doesn’t publish a training cutoff, so we count from its release date. The real gap is larger.
How to call it
3 ways to use TeleOCR (ex NaviDC-OCR).
| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| Hugging Face | huggingface.co | ||
| GitHub | github.com | ||
| Web app | teleai.com.cn |
Pricing
Price not published.
No price source linked. Checked 6 October 2026.
Specs
- Input
- Images and PDF
- Output
- Text
- Context window
- Not published
- Max output
- Not published
- Open weights
- Yes
- Licence
- Apache-2.0
- Training cutoff
- Not published
- Released
- 17 August 2026
- Status
- Current
- Type
- Multimodal model
What stands out
What its maker marketed at launch, and what people found later, each with its source.
Camera-captured document parsing without dewarping
One ~1.4B-parameter Qwen2.5-VL-based model parses both digital and photographed or curved documents directly, with no separate rectification step (DocUNet/DIR300 demos).
Dr.DocBench Challenge lead
Found after launch
Self-reported 67.96 overall on the EMNLP 2026 Dr.DocBench Challenge vs MinerU 2.5 Pro 62.26 and PaddleOCR-VL 1.6 55.11.
Notes
Released Aug 17, 2026 as NaviDC-OCR (StarDoc-AI), renamed TeleOCR on Sept 10. ~1,280 HF likes and ~41k downloads by Oct 6, 2026; GGUF/llama.cpp ports exist. Benchmarks are self-reported. The README links TeleAI’s (China Telecom) document-parsing web app; the HF org is XingChen-AGI.