News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
Minerva

Minerva

An Italian-first large language model trained from scratch at Sapienza University of Rome.

Minerva is a large language model built by the Sapienza NLP research group at Sapienza University of Rome, led by Roberto Navigli. Released in April 2024 and trained from scratch with Italian as its main language, it matters as a fully open model whose training data and weights are both publicly available.

Share on WhatsApp

  • Trained from scratch with Italian as its primary language
  • Minerva 7B has 7 billion parameters
  • Roughly 2.5 trillion tokens, half Italian and half English, plus 200 billion tokens of code
  • Trained on CINECA's Leonardo supercomputer
  • Both models and training data are publicly available

What it is

Minerva is a large language model for natural language processing, made by the Sapienza NLP research group at Sapienza University of Rome under Roberto Navigli. It was released in April 2024 and built from scratch with Italian as its primary language, alongside English. Using deep learning and a Transformer design, it reads and produces human-like text, and it has been fine-tuned for tasks such as translation, summarisation and question answering. Its main release, Minerva 7B, carries 7 billion parameters. Both the model and its training data are publicly available.

How it came about

Minerva 7B was developed inside the Future Artificial Intelligence Research (FAIR) project. CINECA took part by providing its Leonardo supercomputer for training, and further contributions came from Babelscape and the CREATIVE PRIN Project. The model appeared in April 2024. In 2026 an extended version of the model, able to handle long documents and visual inputs, was built into a multimodal assistant called ChatMinerva.

Inside the model

Minerva uses a Transformer architecture, the deep learning design that processes and generates text. It was trained from scratch rather than adapted from an existing model. Minerva 7B has 7 billion parameters and saw roughly 2.5 trillion tokens during training, divided evenly between Italian and English text, plus 200 billion tokens of code. Training ran on the Leonardo supercomputer at CINECA. The model was then fine-tuned for language tasks including translation, summarisation and question answering.

What it can do

Minerva handles natural language processing work, understanding and producing human-like text. It has been fine-tuned for language translation, text summarisation and question answering. The 2026 ChatMinerva version extends the model to long documents and visual inputs, and adds web search, stronger safety measures and optical character recognition, the reading of text from images.

How to use it

Minerva is described as truly open: both the models and the training data can be reached by the public. It is offered as a web app, and the project runs a site at minerva-llm.org. ChatMinerva is the assistant built on top of the model.

Safety and testing

ChatMinerva, the 2026 assistant built on an extended Minerva model, includes enhanced safety components alongside web search and optical character recognition.

Why it matters

Minerva stands out as a model trained from scratch for Italian rather than adapted from an English-centred system, and as a fully open release with public data as well as public weights. Its extended form was carried into ChatMinerva, a multimodal assistant combining the model with web search, safety measures and optical character recognition.

Questions people ask about Minerva

What is Minerva?

It is a large language model built by the Sapienza NLP group at Sapienza University of Rome, led by Roberto Navigli. Released in April 2024, it was trained from scratch with Italian as its main language and handles tasks such as translation, summarisation and question answering.

How large is it and what was it trained on?

Minerva 7B has 7 billion parameters. It was trained on about 2.5 trillion tokens, split evenly between Italian and English text, together with 200 billion tokens of code. Training ran on the Leonardo supercomputer provided by CINECA.

Can I use it myself?

Yes. Minerva is described as truly open, with both the models and the training data available to the public, and it runs as a web app through the project site minerva-llm.org.

What is ChatMinerva?

ChatMinerva is a multimodal AI assistant from 2026 that uses an extended Minerva model able to process long documents and visual inputs. It adds web search, enhanced safety measures and optical character recognition on top of the model.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "Minerva (model)", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryLLM rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.