NuMind

NuMind builds vision language models that turn documents into structured JSON and clean Markdown. Its flagship model family, NuExtract, handles two jobs: structured information extraction from files such as forms, contracts and claims, and OCR style content extraction. The company runs these models inside its own SaaS platform and also deploys them privately for customers. The product is aimed at document heavy industries, with insurance, banking and healthcare named on the site. A team defines the fields it wants, sends in documents, and receives structured output it can pass to downstream systems, which cuts the manual review that generic extraction workflows usually need. The vendor cites a case study in which an insurer moved field level accuracy from 30 percent to 97 percent after switching to NuExtract, and it reports that NuExtract3 scores 65.2 on its structured extraction benchmark against 53.8 for the next best model of similar size. NuMind also publishes open source models on Hugging Face and research papers, including NuNER, an entity recognition encoder pretrained with LLM annotated data. Teams that want to run the models themselves can use the open releases, while the SaaS platform and private enterprise deployment cover managed use. The vendor does not publish prices. The pricing page returns the home page, so cost for the SaaS platform or a private deployment is not listed and has to be arranged with the company.