Latam-GPT: Latin American Artificial Intelligence
A pioneering project aims to create the first language model designed from and for Latin America and the Caribbean, with a focus on digital sovereignty, inclusion and respect for regional identity. Artificial intelligence is advancing at great speed, but many of the technologies that now shape our interaction with the digital world…

A pioneering project aims to create the first language model designed from and for Latin America and the Caribbean, with a focus on digital sovereignty, inclusion and respect for regional identity.
Artificial intelligence is advancing at great speed, but many of the technologies that shape how we interact with the digital world today were developed with a vision centered on the global north.
This approach has left out vast regions such as Latin America, producing models that do not understand our languages, cultures or social realities. Against this gap, Latam-GPT emerges, an initiative coordinated by Chile's National Center for Artificial Intelligence (CENIA), which seeks to change the rules of the game.
A Latin American response to technological bias
Latam-GPT is born as a large language model (LLM) designed specifically for the cultural, linguistic and social particularities of Latin America and the Caribbean. Its development responds to multiple limitations detected in models such as ChatGPT or Gemini, trained on data that is predominantly in English and from foreign cultural perspectives.
Among the most notable problems are “hallucinations” —incorrect or invented answers— about Latin American topics. As researcher Alexandra García points out, when these models answer questions about the region's literature or history, “they normally invent titles, authors or make mistakes.” This is due to a structural lack of local data during training.
In addition, many current commercial models resort to opaque or even predatory data collection practices. According to Latam-GPT's backers, it is necessary to move toward an ethical, transparent and participatory approach to developing these technologies.
Digital sovereignty with a local face
The heart of Latam-GPT is its commitment to digital sovereignty. For Rodrigo Durán Rojas of CENIA, the goal is clear: “This model allows Latin America's culture, idiosyncrasy and worldview to be embodied in a transformative technology.”
The project also distances itself from the closed, proprietary logic of the big commercial models. Its development will be open and collaborative, allowing more people and institutions in the region to study, use and improve the tool. This openness democratizes knowledge and encourages building local capacities.
A regional ecosystem under way
Latam-GPT is the result of a broad collaboration network: more than 30 institutions and 60 experts from countries such as Chile, Mexico, Colombia, Uruguay, Peru, Argentina, Costa Rica, and even Spain and the U.S. are actively taking part. Partners include universities, libraries, foundations, government bodies and civil society organizations.
From the Dominican Republic, the channel CDN Canal 37 stands out as the country's first private company to contribute data to a massive language model, marking a milestone for Caribbean participation in AI projects.
In addition, the Dominican company Tabuga SRL and Chile's National Center for Artificial Intelligence (CENIA) have formed a strategic alliance to promote artificial intelligence in the Dominican Republic and the Caribbean. Tabuga will act as CENIA's official representative in the region, promoting development services and cutting-edge AI solutions for businesses. Among the most relevant joint initiatives is the training program “AI for Everyone”, which aims to teach professionals about the potential of artificial intelligence, as well as the joint development of Latam-GPT, a model trained in Spanish to reflect Latin American culture and context, thereby strengthening regional technological sovereignty.
The technical development relies on a supercomputing infrastructure at the Universidad de Tarapacá (Chile), with a planned investment of USD 10 million and a cluster of NVIDIA H200 GPUs. This capacity will make it possible to train a model with 50 billion parameters, comparable to GPT-3.5.
The initiative also introduces innovative tools such as CopuChat, a conversational interface that allows citizens to contribute data anonymously to train the model, strengthening transparency and community participation.
Funding and institutional backing
The project has the backing of Chile's Ministry of Science, Technology, Knowledge and Innovation, and has received an initial investment of US$250,000 from CAF (Development Bank of Latin America and the Caribbean), a key step toward hiring talent and starting to train the model.
Challenges ahead
One of the main challenges is the legal framework around copyright. Strict laws in many countries of the region make it hard to use cultural works for scientific purposes, unlike in the U.S., where large companies can afford to operate amid legal uncertainty. The Latam-GPT team advocates data collection that respects privacy, avoids unwanted commercial uses and is based on non-profit or research purposes.
What comes next
The first working version of Latam-GPT is expected by mid-2025, with a clear focus on continuous improvement and expanding alliances. The final goal is not only technical but deeply social: that every community in the region understand what artificial intelligence is and how it works, and be able to decide whether, how and with which tools it wishes to interact with it.
Latam-GPT is more than a language model. It is an act of cultural affirmation, a strategy for technological autonomy and an invitation to build a digital future in which Latin America is not just a user, but a protagonist.