Github: OpenBMB/MiniCPM-OCR
MiniCPM-OCR 1.0：huggingface.co
MiniCPM-OCR Jev：huggingface.co
7 likes · 1 repost · 308 views
Топ-10 GitHub за 3 дня: tongjiang-lab, MiniCPM-OCR
Обзор: топ-10 новых репозиториев GitHub за 26–29 октября 2026 — open-source проекты, которые быстрее всех набрали звёзды.
В выпуске: 54 офлайн-3D-урока по школьным предметам; OCR-модель MiniCPM с проверкой качества; океан на three.js с волнами FFT.
1. smithjohn78390/tongjiang-lab — 54 интерактивных 3D-урока по физике, химии, биологии и географии (HTML, +742 ★)
2. OpenBMB/MiniCPM-OCR — OCR-модель MiniCPM-OCR 1.0 на 1,3 млрд параметров и оценщик Jev (Python, +405 ★)
3. MiniMax-AI/mcode-stack — Набор скиллов и рабочих правил команды MiniMax Code для ИИ-агентов (Python, +359 ★)
4. haydenschmitty/floorplan — Плагин для Claude Code и Codex: интерактивная карта кодовой базы (JavaScript, +314 ★)
5. toyxyz/toyxyz_RGBMocap — Захват движения с видео в Windows: тело и кости в 3D локально (Python, +267 ★)
6. brainstormity/LightPost — Самостоятельный планировщик постов в соцсети с одного дашборда (Python, +217 ★)
7. Humpalumps/Ultra-Water — Океан в реальном времени для three.js: волны FFT, буруны, пена, каустика (JavaScript, +195 ★)
8. ngxson/esp32-usb-to-airprint — ESP32-S3 превращает старый USB-принтер в сетевой с AirPrint и сканером (C, +177 ★)
9. molence/Mol-1 — Открытая модель предсказания структур белков, ДНК, РНК и лигандов (Python, +173 ★)
10. oxnw/ai-phone — SIP-телефон с локальными речевыми моделями на Apple Silicon (Python, +160 ★)
Источник: GitHub Search API, репозитории созданы с 2026-10-26, отсортированы по звёздам.
Лидер: smithjohn78390/tongjiang-lab
Остальные: OpenBMB/MiniCPM-OCR, MiniMax-AI/mcode-stack, haydenschmitty/floorplan, toyxyz/toyxyz_RGBMocap, brainstormity/LightPost, Humpalumps/Ultra-Water, ngxson/esp32-usb-to-airprint, molence/Mol-1, oxnw/ai-phone
#GitHub #opensource #программирование
Каждые 3 дня — 10 новых open-source проектов GitHub, быстрее всех набравших звёзды. Подписывайся, чтобы не пропустить следующий выпуск.
1 like · 6 views
Free AI Reader Turns Messy Scans Into Text And Flags Its Own Mistakes
Open-source AI lab OpenBMB has released MiniCPM OCR, a free tool that turns scans, phone photos and PDFs into text and marks the spots it may have misread.
Key Takeaways
- OpenBMB released MiniCPM OCR, a 1.3 billion-parameter OCR model family for PDFs, scans and photographed documents
- The GitHub repository says the main model scored 93.68 on Real5-OmniDocBench and improved end-to-end processing efficiency by 55%
- The repository says the system handles Chinese and English documents, tables, formulas, charts and key information extraction
- The GitHub page says the model runs about 1.49 times faster than PaddleOCR-VL-1.6 and 3.62 times faster than MinerU2.5 Pro
OpenBMB posted the launch on X and linked a GitHub repository describing MiniCPM OCR 1.0, a suite for text recognition, document parsing and error review. The repo says the main model scored 93.68 on Real5-OmniDocBench and improved end-to-end processing efficiency by 55%.
OpenBMB builds open-source multimodal models that mix text and vision. OCR, short for optical character recognition, converts text in images or scanned files into machine-readable content.
MiniCPM OCR Targets The Boring Work AI Still Misses
Document parsing remains one of AI’s most expensive failure points.
Models can answer hard questions in chat yet still miss a seal, a table cell or a number on a warped invoice. The repository says the system handles Chinese and English documents, tables, formulas, charts and key information extraction.
It also includes an evaluation component that flags suspect regions rather than treating every output as equally trustworthy.
That region-level confidence is crucial in financial, legal and health workflows where users need to know which part of a page still requires human review, not just that text was extracted.
Also Read: Semantic Scholar Lab Unleashes Free Tool That Finds Papers Searches Miss
The New Release Follows A Broader OpenBMB Push
The launch is separate from OpenBMB research released Monday about human nudge performance in agentic tasks, which examined model decision making. This release is a product move into document ingestion, a different layer of the stack.
The timing follows a wider industry pattern, with multimodal labs taking on front-end tasks once left to specialist software, from coding and support to speech and paperwork.
OCR feeds retrieval systems many AI agents depend on, if a page is read badly, downstream steps inherit mistakes, which is why even strong large language models can fail on messy real-world files.
Where The Speed Claims Matter Most
The GitHub page says the model runs about 1.49 times faster than PaddleOCR-VL-1.6 and 3.62 times faster than MinerU2.5 Pro under the listed decoding setup. Those comparisons matter where thousands of pages hit a queue every hour.
Faster processing determines whether a bank, insurer or business process outsourcer can review documents in near real time or needs overnight batches and exception teams.
Benchmarks are useful, but production OCR often breaks on ugly inputs, phone photos, curved pages and low-contrast scans. OpenBMB says the system was trained for those cases.
Real-world users will test whether those claims hold outside curated datasets.
How Confidence Scoring Changes OCR Review
The repository describes MiniCPM-OCR Jev, which evaluates results at the region level and marks areas that need review. Traditional tools often return one final text output and force staff to hunt for mistakes line by line.
A region-level checker narrows the search by pointing reviewers to the exact table, stamp or paragraph likely to be wrong.
For operations teams, that can cut review cost without pretending the model is perfect. A bank processing mortgage files benefits from something that parses most pages quickly and highlights the small share of risky fields before they hit underwriting or compliance systems.
Read Next: GitHub Gives All Copilot Users OpenAI Tool That Solves More Code
---
Originally published on Fathom News.
Hugging Face:
MiniCPM-OCR 1.0: huggingface.co
MiniCPM-OCR Jev: huggingface.co
GitHub: OpenBMB/MiniCPM-OCR
Try the demo: platform.modelbest.ai
Use the API: platform.modelbest.ai
11 likes · 2 reposts · 411 views
OCR never tells you which line it got wrong.. so you re-read the whole page anyway.
@OpenBMB tackles that. MiniCPM-OCR 1.0 (1.3B) parses 70+ pages a minute, and MiniCPM-OCR Jev flags the outputs that need a human check.
Would you trust it?
🔗 GitHub: OpenBMB/MiniCPM-OCR
🤗 Hugging Face:MiniCPM-OCR 1.0: huggingface.co
42 likes · 10 reposts · 1,877 views
Hugging Face:
MiniCPM-OCR 1.0: huggingface.co
MiniCPM-OCR Jev: huggingface.co
GitHub: OpenBMB/MiniCPM-OCR
Try the demo: platform.modelbest.ai
@OpenBMB
6 likes · 951 views
Hugging Face:
MiniCPM-OCR 1.0: huggingface.co
MiniCPM-OCR Jev: huggingface.co
GitHub: OpenBMB/MiniCPM-OCR
Try the demo: platform.modelbest.ai
Use the API: platform.modelbest.ai
6 likes · 475 views
OpenBMB released a 1.3B OCR model and a separate 2B model to check its work
A PDF turning into clean Markdown doesn't mean all the numbers survived the trip.
OpenBMB just released MiniCPM-OCR 1.0, a 1.3B model for parsing Chinese and English documents, alongside MiniCPM-OCR Jev, a 2B model for checking OCR output.
Jev takes an image crop and its corresponding OCR result, then returns a quality score from 0 to 1. It handles text, formulas and tables. It can also score results from other OCR parsers, so using their parser isn't a requirement.
That's the part I'd want in a local RAG pipeline. Feed it a table with a wrong number and your chatbot can very helpfully cite that wrong number back to you. I'd use something like this to flag questionable regions for review before they go into the index.
The code and both models' weights are Apache 2.0. The GitHub repo has local deployment instructions and links to quantized weights.
30 points · 3 comments
Mentions
Github: OpenBMB/MiniCPM-OCR MiniCPM-OCR 1.0：huggingface.co MiniCPM-OCR Jev：huggingface.co
Топ-10 GitHub за 3 дня: tongjiang-lab, MiniCPM-OCR Обзор: топ-10 новых репозиториев GitHub за 26–29 октября 2026 — open-source проекты, которые быстрее всех набрали звёзды. В выпуске: 54 офлайн-3D-урока по школьным предметам; OCR-модель MiniCPM с проверкой качества; океан на three.js с волнами FFT. 1. smithjohn78390/tongjiang-lab — 54 интерактивных 3D-урока по физике, химии, биологии и географии (HTML, +742 ★) 2. OpenBMB/MiniCPM-OCR — OCR-модель MiniCPM-OCR 1.0 на 1,3 млрд параметров и оценщик Jev (Python, +405 ★) 3. MiniMax-AI/mcode-stack — Набор скиллов и рабочих правил команды MiniMax Code для ИИ-агентов (Python, +359 ★) 4. haydenschmitty/floorplan — Плагин для Claude Code и Codex: интерактивная карта кодовой базы (JavaScript, +314 ★) 5. toyxyz/toyxyz_RGBMocap — Захват движения с видео в Windows: тело и кости в 3D локально (Python, +267 ★) 6. brainstormity/LightPost — Самостоятельный планировщик постов в соцсети с одного дашборда (Python, +217 ★) 7. Humpalumps/Ultra-Water — Океан в реальном времени для three.js: волны FFT, буруны, пена, каустика (JavaScript, +195 ★) 8. ngxson/esp32-usb-to-airprint — ESP32-S3 превращает старый USB-принтер в сетевой с AirPrint и сканером (C, +177 ★) 9. molence/Mol-1 — Открытая модель предсказания структур белков, ДНК, РНК и лигандов (Python, +173 ★) 10. oxnw/ai-phone — SIP-телефон с локальными речевыми моделями на Apple Silicon (Python, +160 ★) Источник: GitHub Search API, репозитории созданы с 2026-10-26, отсортированы по звёздам. Лидер: smithjohn78390/tongjiang-lab Остальные: OpenBMB/MiniCPM-OCR, MiniMax-AI/mcode-stack, haydenschmitty/floorplan, toyxyz/toyxyz_RGBMocap, brainstormity/LightPost, Humpalumps/Ultra-Water, ngxson/esp32-usb-to-airprint, molence/Mol-1, oxnw/ai-phone #GitHub #opensource #программирование Каждые 3 дня — 10 новых open-source проектов GitHub, быстрее всех набравших звёзды. Подписывайся, чтобы не пропустить следующий выпуск.
Free AI Reader Turns Messy Scans Into Text And Flags Its Own Mistakes Open-source AI lab OpenBMB has released MiniCPM OCR, a free tool that turns scans, phone photos and PDFs into text and marks the spots it may have misread. Key Takeaways - OpenBMB released MiniCPM OCR, a 1.3 billion-parameter OCR model family for PDFs, scans and photographed documents - The GitHub repository says the main model scored 93.68 on Real5-OmniDocBench and improved end-to-end processing efficiency by 55% - The repository says the system handles Chinese and English documents, tables, formulas, charts and key information extraction - The GitHub page says the model runs about 1.49 times faster than PaddleOCR-VL-1.6 and 3.62 times faster than MinerU2.5 Pro OpenBMB posted the launch on X and linked a GitHub repository describing MiniCPM OCR 1.0, a suite for text recognition, document parsing and error review. The repo says the main model scored 93.68 on Real5-OmniDocBench and improved end-to-end processing efficiency by 55%. OpenBMB builds open-source multimodal models that mix text and vision. OCR, short for optical character recognition, converts text in images or scanned files into machine-readable content. MiniCPM OCR Targets The Boring Work AI Still Misses Document parsing remains one of AI’s most expensive failure points. Models can answer hard questions in chat yet still miss a seal, a table cell or a number on a warped invoice. The repository says the system handles Chinese and English documents, tables, formulas, charts and key information extraction. It also includes an evaluation component that flags suspect regions rather than treating every output as equally trustworthy. That region-level confidence is crucial in financial, legal and health workflows where users need to know which part of a page still requires human review, not just that text was extracted. Also Read: Semantic Scholar Lab Unleashes Free Tool That Finds Papers Searches Miss The New Release Follows A Broader OpenBMB Push The launch is separate from OpenBMB research released Monday about human nudge performance in agentic tasks, which examined model decision making. This release is a product move into document ingestion, a different layer of the stack. The timing follows a wider industry pattern, with multimodal labs taking on front-end tasks once left to specialist software, from coding and support to speech and paperwork. OCR feeds retrieval systems many AI agents depend on, if a page is read badly, downstream steps inherit mistakes, which is why even strong large language models can fail on messy real-world files. Where The Speed Claims Matter Most The GitHub page says the model runs about 1.49 times faster than PaddleOCR-VL-1.6 and 3.62 times faster than MinerU2.5 Pro under the listed decoding setup. Those comparisons matter where thousands of pages hit a queue every hour. Faster processing determines whether a bank, insurer or business process outsourcer can review documents in near real time or needs overnight batches and exception teams. Benchmarks are useful, but production OCR often breaks on ugly inputs, phone photos, curved pages and low-contrast scans. OpenBMB says the system was trained for those cases. Real-world users will test whether those claims hold outside curated datasets. How Confidence Scoring Changes OCR Review The repository describes MiniCPM-OCR Jev, which evaluates results at the region level and marks areas that need review. Traditional tools often return one final text output and force staff to hunt for mistakes line by line. A region-level checker narrows the search by pointing reviewers to the exact table, stamp or paragraph likely to be wrong. For operations teams, that can cut review cost without pretending the model is perfect. A bank processing mortgage files benefits from something that parses most pages quickly and highlights the small share of risky fields before they hit underwriting or compliance systems. Read Next: GitHub Gives All Copilot Users OpenAI Tool That Solves More Code --- Originally published on Fathom News.
Hugging Face: MiniCPM-OCR 1.0: huggingface.co MiniCPM-OCR Jev: huggingface.co GitHub: OpenBMB/MiniCPM-OCR Try the demo: platform.modelbest.ai Use the API: platform.modelbest.ai
OCR never tells you which line it got wrong.. so you re-read the whole page anyway. @OpenBMB tackles that. MiniCPM-OCR 1.0 (1.3B) parses 70+ pages a minute, and MiniCPM-OCR Jev flags the outputs that need a human check. Would you trust it? 🔗 GitHub: OpenBMB/MiniCPM-OCR 🤗 Hugging Face:MiniCPM-OCR 1.0: huggingface.co
Hugging Face: MiniCPM-OCR 1.0: huggingface.co MiniCPM-OCR Jev: huggingface.co GitHub: OpenBMB/MiniCPM-OCR Try the demo: platform.modelbest.ai @OpenBMB
Hugging Face: MiniCPM-OCR 1.0: huggingface.co MiniCPM-OCR Jev: huggingface.co GitHub: OpenBMB/MiniCPM-OCR Try the demo: platform.modelbest.ai Use the API: platform.modelbest.ai
OpenBMB released a 1.3B OCR model and a separate 2B model to check its work A PDF turning into clean Markdown doesn't mean all the numbers survived the trip. OpenBMB just released MiniCPM-OCR 1.0, a 1.3B model for parsing Chinese and English documents, alongside MiniCPM-OCR Jev, a 2B model for checking OCR output. Jev takes an image crop and its corresponding OCR result, then returns a quality score from 0 to 1. It handles text, formulas and tables. It can also score results from other OCR parsers, so using their parser isn't a requirement. That's the part I'd want in a local RAG pipeline. Feed it a table with a wrong number and your chatbot can very helpfully cite that wrong number back to you. I'd use something like this to flag questionable regions for review before they go into the index. The code and both models' weights are Apache 2.0. The GitHub repo has local deployment instructions and links to quantized weights.