OCR Models logopeterindia.net
Knowledge Hub

Multimodal Language Models & OCR Engines for Document AI

A directory of OCR engines, vision-language models and document AI services — open-source engines, document-to-markdown converters, multimodal models and cloud services for document processing and agentic workflows.

29 topics•5 categories•Updated October 2026

Open-source OCR Engines

4 entries

Established libraries for text recognition.

01Tesseract OCRTesseract 4 adds a new neural net (LSTM) based OCR engine focused on line recognition, while still supporting the legacy Tesseract 3 engine that works by recognizing character patterns. GitHub lists 5.5.3 as the latest release, from December 2025. 02PaddleOCRUses cutting-edge algorithms and proven performance in real-world applications, already powering popular open-source projects like Umi-OCR, OmniParser, MinerU, and RAGFlow. 03EasyOCRReady-to-use OCR library supporting 80+ languages. GitHub lists v1.7.2 as the latest release, from September 2023, so check its activity before relying on it. 04docTRDocument text recognition library built on TensorFlow and PyTorch. GitHub lists v1.1.0 as the latest release, from July 2026.

Document-to-Markdown & Parsing Tools

5 entries

Convert complex documents into LLM-ready text.

05MinerUTransforms complex documents like PDFs and Office docs into LLM-ready markdown/JSON for your agentic workflows. 06DoclingDocument conversion toolkit from IBM that prepares documents for generative AI. GitHub lists v2.142.1 as the latest release, from October 2026. 07MarkerConverts PDFs and other documents to markdown and JSON. 08olmOCRAi2's toolkit for converting PDFs into clean text with vision-language models. GitHub lists v0.4.27 as the latest release, from January 2026. 09UnstructuredOpen-source library for preprocessing documents for LLM pipelines. GitHub lists 0.27.34 as the latest release, from October 2026.

Vision-language OCR Models

10 entries

Multimodal models with strong document understanding.

10GLM-OCRA multimodal OCR model for complex document understanding, built on the GLM-V encoder–decoder architecture. Introduces Multi-Token Prediction (MTP) loss and stable full-task reinforcement learning to improve training efficiency, recognition accuracy, and generalization. GitHub lists v0.1.5 as the latest release, from March 2026. 11DeepSeek-OCRA two-stage transformer-based document AI that compresses page images into compact vision tokens before decoding them with a high-capacity mixture-of-experts language model. Stage 1 merges a windowed SAM vision transformer with a dense CLIP-Large encoder and a 16× convolutional compressor; Stage 2 uses the DeepSeek-3B-MoE decoder (~570M active parameters per token) to reconstruct text, HTML, and figure annotations with minimal loss. 12Qwen3.8-27B (Alibaba Qwen)Open-weight multimodal model (Apache 2.0, 27B parameters) that handles OCR and document parsing. Its model card reports 91.1 on OmniDocBench 1.5, a vendor-reported score; third-party document benchmarks such as PureDocBench give lower figures, so test it on your own documents. 13MiniCPM-OCR 1.0 (OpenBMB)OpenBMB's first OCR-specific model, 1.3B parameters on a MiniCPM-V 4.6 base, Apache 2.0. The model card reports 93.68 on Real5-OmniDocBench and 54.7 on OCRBench v2 (vendor-reported), for Chinese and English documents, with roughly 1.2 pages per second on an RTX 5090. 14MiniCPM-OCR Jev (OpenBMB)A 2B correctness model that scores a candidate OCR result for a text, formula or table region from 0 to 1, so reliable results can be accepted automatically and the rest sent to human review. It works with output from other parsers; Apache 2.0. 15Unlimited OCR 2 (Baidu)Open-source long-document parser released on 27 October 2026 under the MIT licence. It uses Sparse Reference Sliding Window Attention to keep memory roughly constant on documents of hundreds of pages; Baidu reports state-of-the-art results on MPDocBench and MarathonDocBench. Run it with Transformers or SGLang. 16Mistral OCRDigitize PDFs, scans, DOCX, PPTX, and handwritten sources. Extract to structured JSON with custom templates, parse forms, classify documents, and process images down to charts, signatures, and fine print. 17Qwen (Alibaba Cloud Model Studio)Qwen models on Alibaba Cloud Model Studio, with better OCR across more languages and complex scenes. 18Gemini 3.1 Flash-LiteA low-latency, cost-effective multimodal model optimized for high-frequency, lightweight tasks. Supports text, image, video, audio, and PDF inputs, and is designed for high-volume agentic workflows, simple data extraction, and applications where latency and API cost are the primary constraints. 19Vision Language Models (VLMs)Open and proprietary vision language models, vision encoders, benchmarks and fine-tuning tools.

Cloud Document AI Services

3 entries

Managed services for extraction and understanding.

20Azure Document IntelligenceMicrosoft's service for extracting text, tables and structure from documents. 21Google Document AIGoogle Cloud service for document processing and extraction. 22Amazon TextractAWS service that extracts text, forms and tables from scanned documents.

Related Pages

7 entries

Related directories on this site.

23Multimodal Language ModelsModels that combine text and vision. 24Computer VisionComputer vision hub. 25Large Language ModelsLanguage models directory. 26Natural Language ProcessingNLP hub. 27Data Science PlatformsData science platforms. 28Generative AIGenerative AI hub. 29Multimodal Embedding ModelsMultimodal embedders for images, video, audio and documents.

Frequently Asked Questions

What is OCR?

Optical character recognition converts images of text, such as scans and photos, into machine-readable text.

What is document AI?

Document AI goes beyond text to understand layout, tables and forms, producing structured output such as markdown or JSON for downstream systems.

Classic OCR engine or a vision-language model?

Engines such as Tesseract and PaddleOCR are fast and cheap for clean text; vision-language models handle complex layouts, handwriting and reasoning but cost more compute.

Which tools prepare documents for LLMs?

MinerU, Docling, Marker, olmOCR and Unstructured convert PDFs and Office files into markdown or JSON for retrieval and agent workflows.

Why were Boost.ai and the DeepSeek-OCR site removed?

Boost.ai is a conversational AI platform rather than OCR, and the deepseek-ocr.io address now returns an error, so the card links to the project's GitHub repository instead.

Which OCR models are new in October 2026?

Open models released around late October 2026 include Baidu Unlimited OCR 2 (long-document parsing, MIT licence), OpenBMB MiniCPM-OCR 1.0 (1.3B parameters) with its MiniCPM-OCR Jev correctness checker (2B), and the general-purpose Qwen3.8-27B, which reports 91.1 on OmniDocBench 1.5. These scores are vendor-reported, so test on your own documents.