🚨炸了！阿里 ATH-MaaS 发布 OvisOCR3：仅 0.8B 的统一小模型，一个 checkpoint 干翻整套文档处理流水线！
This is a dev post classified by AGK Intelligence as AI dev tools (a launch), kept by the Dev Radar because it carries real work, not commentary.
🚨炸了！阿里 ATH-MaaS 发布 OvisOCR3：仅 0.8B 的统一小模型，一个 checkpoint 干翻整套文档处理流水线！ 同尺寸SOTA硬核性能： 🔹OmniDocBench 96.74（逼近顶级专用模型） 🔹PureDocBench 82.31 第一，超 Qwen3.8-Max 🔹OCRBench v2 英/中双榜第一（65.9/61.3） 🔹KIE 全面领先，跨页表格 Relation F1 97.07 反超大模型 主要亮点： 1️⃣文档解析 → Markdown/JSON（LaTeX公式+HTML表格） 2️⃣通用OCR + 边界框 + 图表数据恢复 3️⃣关键信息抽取（按字段/Schema直接出JSON） 4️⃣图表定位+Caption + 流程图转Mermaid 5️⃣跨页表格智能重建（自动判断并拼接） 基于 Qwen3.5-0.8B + 多阶段蒸馏/RL，Prompt 一键切换任务，Apache-2.0 开源，1.7GB 级显存本地可跑，RAG/企业文档管线直接能上。 小模型真的能替代复杂流水线了！ ModelScope：https://modelscope.ai/models/ATH-MaaS/OvisOCR3 HF：https://huggingface.co/ATH-MaaS/OvisOCR3 报告：https://arxiv.o
Posted by SuSu_酥酥👅 (24.7k followers) 2 days ago · 99 likes · 32.2k views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: Step-5-Preview, ovisocr3, OvisOCR3 Technical Report.
More dev work like this
- 🎃Dario给中国人民带来的极致恐怖从Claude无情封号开始 — @NFT_Chen
- Mia's Qwen3.8-Flash-Next recipe on a DGX Spark: one command, first try, 2 min to the… — @niklaslenz_ai
- 给AI喂一整仓，最烦的就是它每回都从头翻文件，这毛病你多半也遇到过。 — @bkdgiffug
- Coding agents aren't slow because they're dumb. They're slow because every edit waits on… — @RunAnywhereAI
- ICYMI: Foursquare co-founder @dens sat down with our Head of Product, AI, Jeff Smith at… — @automattic
- 🤯惊了！9B参数大模型，只需16G内存即可本地跑通，能力反超Qwen3.5-27B！ — @NFT_Chen
- Introducing JetSpec++: we find causal parallel drafting paired with hidden states… — @haoailab
- Build the agent, test the agent, grade the results, change the agent, run the test cases… — @rseroter
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.2k posts from 5k X accounts over the last 21 days, 8.8k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-10-30 20:27 UTC. Full method.