Chile launches open-source AI model designed for Latin America

BUENOS AIRES, Argentina — Chile on Tuesday launched the first open-source artificial intelligence language model trained on the diverse cultures of Latin America, aiming to better reflect regional realities and strengthen the region's presence in the global AI race.

Latam-GPT is the result of a two-year regional effort led by the National Center of Artificial Intelligence of Chile, CENIA, and supported by over 30 institutions across eight Latin American countries.

''Artificial intelligence is the greatest technological revolution of recent times, and from Latin America and the Caribbean, it is strategic and urgent that we play a role,'' Chilean President Gabriel Boric said Tuesday after the launch, noting that the new system will be key to adding Latin American data and identity to AI.

Announced at the February 2025 Artificial Intelligence Action Summit in Paris, the project launched in early 2023 to address linguistic biases in models trained primarily on English data. Rather than competing directly with consumer tools like ChatGPT or Google Gemini, Latam-GPT acts as a foundational infrastructure for future regional applications.

''Latam-GPT is trained with a proportion of Latin American data that previously did not exist online and was not included in existing models,'' said Rodrigo Durán, executive director at CENIA. ''This allows for more accurate, correct and efficient performance when it comes to Latin America and the Caribbean.''

Latam-GPT is trained on data from private sources obtained through strategic partnerships across the region, as well as synthetic data used to address areas identified as underrepresented, said Gabriela Arriagada, a researcher at CENIA and head of the project's ethics team. Developing Latam-GPT required collecting more than eight terabytes of data, equivalent to millions of books.

''When we talk about incorporating Latin American culture, we are referring to a training approach designed to address data that reflects cultural realities, identifying where gaps exist in other models, understanding their shortcomings, and gradually building knowledge to improve that representation,'' Arriagada added.

''That provides some assurance that, at the very least, everyone is included in the training,'' Chiruzzo added.

For now, the project will operate primarily in Spanish and Portuguese, with plans to incorporate Indigenous languages in later stages.

The development of Latam-GPT means the region now has the technical capacity to build AI models, according to Durán, the executive director at CENIA.

''The fact that Latin America has come together to form a collaborative group is a very positive sign. It shows that Latin America can develop and understand how to create this technology, which also has important implications for regulation, because you cannot regulate something you do not understand,'' he added.

Chiruzzo believes it will be difficult for Latam-GPT to compete with larger corporations with greater resources.

The race for AI leadership has led countries to rethink their policies and initiatives for developing AI technologies. The United States, China and the European Union have more than half the world's most powerful data centers to develop and support AI systems, according to data published by Oxford University. Africa and South America have almost no AI hubs, according to the report.

Chile has been racing to expand its role in the AI boom over the past few years by attracting new talent and building new data centers. In June last year, President Boric said in his State of the Union speech that the country must embrace AI, adding that a ''country that does not invest in artificial intelligence risks falling behind in tomorrow's global landscape.''

Latam-GPT was developed with just $550,000 in funding from CENIA's budget and the Development Bank of Latin America (CAF). The team used Amazon Web Services' cloud to develop its first version, which will be launched at the end of February. The subsequent versions will be trained on a supercomputer at the University of Tarapacá in northern Chile, which costs about $4.5 million, starting in the first semester of 2026.

___