document-intelligence

Datalab 发布 Indra OCR——一个仅约 1M 参数的轻量 OCR 模型，在 CPU 上每秒可处理多达 49 页，速度比 Tesseract 和 PaddleOCR 快 11-24 倍，可将 PDF 和图片转换为 Markdown、纯文本或带框的 JSON。

Datalab released 'lift', an open-weights 9B vision model that extracts schema-constrained JSON directly from PDFs and images in a single step, matching Gemini Flash 3.5 accuracy at 3x the speed. It handles multi-page documents, is self-hostable via vLLM or Hugging Face, and is available under Apache-2.0 code / OpenRAIL-M model licenses.

Jina-ocr-v1 is a multimodal vision language model designed for advanced document intelligence, reading text from images in multiple languages.

Datalab announces Surya OCR 2, a 650M parameter OCR model that achieves top accuracy under 1B parameters on olmOCR benchmarks, with high speed and multilingual support.

Chunkr is an open-source document intelligence service that converts PDFs, PPTs, Word docs, and images into structured chunks for RAG and LLM pipelines. It features layout analysis with OCR, structured HTML/Markdown output, vision-language model processing, and self-hosted deployment via Docker Compose with configurable LLM providers.

Mistral AI releases Mistral OCR 4, a compact document intelligence model that provides bounding boxes, block classification, and inline confidence scores for structured text extraction. It supports 170 languages, runs in a single container for self-hosted deployment, and integrates with the Mistral Search Toolkit for enterprise search and RAG pipelines.

Notum is an AI-powered research and document intelligence tool designed specifically for law firms, aiming to streamline legal research and document management.

Parse-Flow is an open-source visual workflow designer that composes document intelligence primitives (parsing, extraction, classification, splitting) into reusable pipelines, backed by LlamaIndex and a Python worker.

Parse-Flow is an open-source visual workflow designer built by LlamaIndex that chains four document processing primitives—Parse, Classify, Split, and Extract—into a drag-and-drop canvas powered by LlamaAgents workflows, enabling reliable structured data extraction from unstructured enterprise documents like PDFs, contracts, and invoices.

Surya OCR is a state-of-the-art open-source OCR model with less than 1B parameters, supporting 91 languages and achieving top benchmark scores under 3B params.

This paper identifies Footprint Bias in document layout analysis robustness evaluation and proposes a structure-aware auditing framework that decouples probe construction and pathway attribution, showing that small structurally targeted probes cause comparable downstream degradation to larger perturbations.