← 回總覽

智谱 AI 发布 GLM-OCR:10 亿参数以下规模实现高性能

📅 2026-03-16 02:16 Jerry Liu 人工智能 3 分鐘 3650 字 評分: 84
GLM-OCR 智谱 AI OCR 文档解析 小型语言模型
📌 一句话摘要 智谱 AI 全新 GLM-OCR 模型在 OmniDocBench V1.5 上以 0.9B 参数量获得 94.62 的最高分。 📝 详细摘要 Jerry Liu 强调了智谱 AI 发布 GLM-OCR 技术报告的消息,并赞扬了其高效性。这款模型在 OmniDocBench V1.5 排行榜上以 94.62 分位居榜首,尽管其参数量仅为 0.9B。推文指出,小型文档解析模型(如 GLM-OCR、dots.ocr 和 Deepseek)正迅速变得非常出色,这一显著趋势有望实现高质量、本地化、低成本的文档处理。 📊 文章信息 AI 评分:84 来源:Jerry Liu(@je
Skip to main content ![Image 1: LogoBestBlogs](https://www.bestblogs.dev/ "BestBlogs.dev")Toggle navigation menu Toggle navigation menuArticlesPodcastsVideosTweetsSourcesNewsletters

⌘K

Change language Switch ThemeSign In

Narrow Mode

Zhipu AI Releases GLM-OCR: High Performance at Sub-1B Scale ===========================================================

Zhipu AI Releases GLM-OCR: High Performance at Sub-1B Scale =========================================================== ![Image 2: Jerry Liu](https://www.bestblogs.dev/en/tweets?sourceId=SOURCE_560a80) ### Jerry Liu

@jerryjliu0

Zhipu AI released the GLM-OCR technical report yesterday. A model that tops on OmniDocBench V1.5 with a 94.62 score - with only 0.9B params!

I give them credit where credit is due: we are genuinely excited about any research that pushes the frontier of document parsing at sub-1B scale.

Between GLM-OCR, dots.ocr, paddleOCR, Deepseek, small doc parsing models are getting really good really quickly 📈

!Image 3: Tweet image

!Image 4: David Hendrickson

#### David Hendrickson

@TeksEdge · 1d ago

🚨 Want to parse complex PDFs with SOTA accuracy, 100% locally? 📄🔍

At just 0.9B parameters, you can drop GLM-OCR straight into LM Studio and run it on almost any machine! 🥔

🧠 0.9B total parameters

💾 Runs on < 1.5GB VRAM (or ~1GB quantized!)

💸 Zero API costs

🔒 Total data privacy

Desktop document AI is officially here. 💻⚡Show More

!Image 5: Tweet image

51

203

2,264

280.5K

Mar 15, 2026, 6:16 PM View on X

10 Replies

13 Retweets

148 Likes

12K Views ![Image 6: Jerry Liu](https://www.bestblogs.dev/en/tweets?sourceid=560a80) Jerry Liu @jerryjliu0

One Sentence Summary

Zhipu AI's new GLM-OCR model achieves a top score of 94.62 on OmniDocBench V1.5 with only 0.9B parameters.

Summary

Jerry Liu highlights the release of Zhipu AI's GLM-OCR technical report, praising its efficiency. The model tops the OmniDocBench V1.5 leaderboard with a score of 94.62 despite its small size (0.9B parameters). The tweet notes a significant trend where small-scale document parsing models (like GLM-OCR, dots.ocr, and DeepSeek) are rapidly improving, enabling high-quality, local, and low-cost document processing.

AI Score

84

Influence Score 38

Published At Yesterday

Language

English

Tags

GLM-OCR

Zhipu AI

OCR

Document Parsing

Small Language Models HomeArticlesPodcastsVideosTweets

Zhipu AI Releases GLM-OCR: High Performance at Sub-1B Sca... ===============

查看原文 → 發佈: 2026-03-16 02:16:23 收錄: 2026-03-16 06:01:06

🤖 問 AI

針對這篇文章提問,AI 會根據文章內容回答。按 Ctrl+Enter 送出。