ChatMinerva is now live: a new evolution of Minerva Large Language Model.
ChatMinerva is a multimodal AI assistant designed to understand text, images, and documents, access the web in real time, and converse in Italian with greater reliability and safety. This is a research project developed entirely in Italy. It builds on an open, transparent, and fully controllable approach across every stage of development: from pretraining to fine-tuning, through to its safety and moderation mechanisms.
From language model to AI assistant
Users can now upload photographs, images, scanned pages, scientific articles, reports, and technical documentation, and ask ChatMinerva to interpret, summarise, analyse, or answer questions about their content.
Key new features include:
- Multimodal understanding: ChatMinerva can now combine visual and textual information, and users can also interact with the model through voice
- Live web access: through a web RAG system powered by the open-source DuckDuckGo search engine, ChatMinerva can answer questions using up-to-date information from the web
- Long document handling: the model’s context window has been extended to 32,000 tokens via continual training, enabling it to process complex documents and extended conversations
- Enhanced safety: a dedicated component monitors and validates both user inputs and system outputs to moderate unwanted, unreliable, or sensitive content
A model trained together with its users
The new version of the Minerva LLM powering the assistant was fine-tuned on millions of instruction examples, both textual and multimodal, made possible by the computational power of CINECA’s Leonardo supercomputer, used across all training processes.
The direct contribution of Minerva’s growing user community was equally decisive: interactions collected during the public use of Minerva 7B significantly improved the system’s conversational capabilities and response quality.
“Minerva has grown alongside its users. Every prompt, interaction, and piece of feedback has helped build an AI assistant that is more useful, more robust, and closer to people’s real needs,” says Roberto Navigli, Sapienza professor and project lead.
Growing Italian AI technology together
Minerva remains Italy’s leading initiative in the field of Large Language Models developed with direct control over training sources and processes.
“We want to show that it is possible to build frontier AI technology in Europe and in Italy, with an open, scientifically rigorous, and independent approach,” says Navigli. “ChatMinerva was built with far more passion than budget, thanks to the relentless work of dozens of researchers, PhD students, and collaborators who believe in the possibility of creating Italian AI technology: one that can serve as a foundation for competitive products, born in a very different context from that of the global tech giants, but powered by the strength of collaboration between Italy’s leading university and a successful spin-off, Babelscape, which has just celebrated 10 years of intense research and industrial innovation.”