Noxtua AI Models
Models as foundation
Discover the Noxtua AI Model family revolutionizing legal workflows. Built on specialized AI models, specifically tailored for the legal domain, Noxtua is the secure, safe, and specialized alternative from Germany, proudly serving as Europe's leading sovereign Legal AI.
Noxtua's advanced AI models are custom-trained on exclusive high-quality legal data sets and ensure professional confidentiality and compliance with GDPR.
The upcoming Beck-Noxtua is trained with exclusive data from Germany’s leading legal publisher C.H.Beck. Their dataset beck-online is the largest legal one in the German-speaking world with well over 55 million documents covering all relevant areas of law. beck-online contains, among other things, the most comprehensive collection of relevant commentary literature, which is essential for lawyers in their daily work.
Powerful AI and legal expertise
Noxtua Legal LLM
The proprietary Noxtua Legal Large Language Model (LLM) is specialized in the legal domain and trained on exclusive high-quality legal data sets selected and meticulously labeled by legal experts. The datasets are among others provided by members of the Legal AI Alliances such as the law firms CMS and Nordemann. They include documents such as various court decisions from Germany and other EU countries, contract templates and interactions with contracts and lawyers as well as a synthetically generated and legally reviewed dataset. Noxtua has no access to their customers' data which is being processed while using the Legal AI.
Boasting a processing capacity of up to 256,000 tokens context length and 111bn parameters, the Noxtua Legal LLM is well-suited to process long legal texts.
Noxtua Voyage Embed
Noxtua Voyage Embed is a fine-tuned legal search embedding model specialized in European and German law, jointly launched by Noxtua, Stanford spin-off Voyage AI and dejure.org, one of Germany’s most extensive legal databases. Noxtua Voyage Embed (voyage-law-2-xayn) outperforms Open AI’s most potent search model (text-embedding-3-large) on average by a factor of 2 on legal text benchmarks, boasting 1.7 times better search accuracy and 2.2 times better ranking quality with a 3x smaller dimensionality. This makes the model at least twice as good at finding relevant legal documents that lawyers are looking for — all while using less energy and resources to run.
The search model has been trained on a vast dataset by dejure.org and Noxtua, comprising approximately 20 billion tokens of legal texts. With a processing capacity of up to 32k tokens context length, Noxtua Voyage Embed proves invaluable for handling lengthy legal texts. This specialized model can be utilized to create custom AI search solutions centered around legal topics, as it has been trained on high-quality legal documents and is thus highly adept in the legal domain.
Noxtua Research
Noxtua Research is based on the Noxtua Research Model (Question-Answer Model), which is fed by the proprietary Noxtua LLM and the search model Noxtua Voyage Embed, which outperforms the most powerful search model from Open AI by a factor of 2 on legal text benchmarks. The specialized search model is a joint development of Noxtua, the Stanford spin-off Voyage AI, and dejure.org, one of the largest German legal databases. Like all AI models in the Noxtua suite, the Noxtua Research Model is trained with exclusive high-quality legal datasets selected and labeled by legal experts specifically for training Noxtua.
