FULL STORY
OvisOCR2: From Launch to Benchmark Lead
Alibaba's Token Hub unveiled OvisOCR2, a lightweight local model for end-to-end OCR and document parsing that can run locally. It later released a technical report as the 0.8B model rose to the top of document parsing benchmarks.
2026-07-13 ~ 2026-07-16 · 2 episodes · 6 posts
Episode 1 · Alibaba Unveils OvisOCR2 for Local Document Parsing (2026-07-13, 4 posts)
Alibaba's Token Hub team released OvisOCR2, an end-to-end OCR and document parsing model based on Qwen3.5-0.8B that can run locally and convert full-page documents directly into structured text. It reportedly scored 96.58 on OmniDocBench v1.6 and also trended on Hugging Face.
- OvisOCR2: A 0.8B Local Document Parsing Model — Sad_External6106 · 2026-07-13
- OvisOCR2 Hits HF Trending — ATH-MaaS · 2026-07-14
- OvisOCR2 Achieves SOTA in Document Parsing — vanstriendaniel · 2026-07-14
- Alibaba Releases OvisOCR2 OCR Model — Xianbao_QIAN · 2026-07-15
Episode 2 · OvisOCR2 Tops Document Parsing Leaderboard (2026-07-15, 2 posts)
OvisOCR2, a 0.8B document parsing VLM fine-tuned from Qwen3.5-0.8B, has topped parsing leaderboards. It aims to convert single-page document images directly into naturally ordered Markdown, covering text, formulas, and tables.
- OvisOCR2 Tops Document Parsing Leaderboard — Littlepharaoh · 2026-07-15
- OvisOCR2: End-to-End Document Parsing Model — Shiyin Lu · 2026-07-16