Tag: #document-ai14 results
Qwen3.5 models (0.8B to 9B) were evaluated on an open document AI benchmark with 9,000+ real documents. The 9B and 4B variants outperform frontier models like Gemini 3.1 Pro and GPT-5.4 in OCR text extraction, VQA, and KIE tasks. However, they lag significantly in table extraction and handwriting OCR.
This research presents a robust microservice architecture for deploying document understanding pipelines at scale. It offers insights into optimizing OCR and LLM-based extraction for high-throughput production environments.
Microsoft will integrate Anthropic's large models into Microsoft 365 Copilot in Word by late May. It provides new smart polishing and document organization options for users. Models operate under Microsoft's security and compliance framework alongside OpenAI options.
Anthropic has released a new Claude extension for Microsoft Word. It allows Claude to read entire documents, edit selected text, process comments, and track changes to maintain user control.
Hugging Face releases Granite 4.0 3B Vision, a compact multimodal model tailored for enterprise documents. It delivers efficient vision-language intelligence for business applications. The model is now available on the Hugging Face platform.
New open IDP Leaderboard evaluates 16 VLMs on 9,000+ documents across three benchmarks: OlmOCR, OmniDoc, and IDP Core. Gemini 3.1 Pro leads narrowly; cheaper variants match flagships except on reasoning tasks. Features a Results Explorer for predictions vs. ground truth.
這項研究提出一套用於建築文件審查的證據導向管線，將文字、幾何、頁面與版本資訊正規化，並以決定性規則進行約束檢查。實驗顯示，增加重疊頁面區塊的解析度在重複測試中提升準確率，但在更廣泛的資料集並未穩定勝過頁面涵蓋率，顯示需要規則感知的證據路由與人工覆核。
This guide demonstrates how to build a server for real-time text extraction from PDF files stored in Amazon S3. It provides a comparative analysis between custom protocol-based extraction and Amazon Textract to help developers choose the right tool.
Google added a 'Notebooks' section to the Gemini AI app for direct NotebookLM access. Users can create and manage notebooks from the sidebar, generate document-based answers, use Canvas for long-form writing, and produce audio overviews all within Gemini. Rollout starts with paid plan users.
Unisound releases U1-OCR, the first industrial-grade document intelligent base large model. This launch marks the beginning of the OCR 3.0 era. It targets robust document processing for industrial applications.