The National Centre of Artificial Intelligence of Chile, CENIA, has been working for over two years, along with over 30 institutions across the region, to develop Latam-GPT.
Chile has unveiled its first-ever homegrown open-source AI language model, which has been trained to incorporate Latin America’s diverse cultures. This model showcases the realities and differences in the region and also elevates its position in the global AI race.
The National Centre of Artificial Intelligence of Chile, CENIA, has been working for over two years, along with over 30 institutions across the region, to develop Latam-GPT. Gabriel Boric, President of Chile, described artificial intelligence as one of the most important technological breakthroughs in recent times, and therefore, it was of utmost importance for LATAM to play its part and contribute to the innovation. He also added that this new system will be instrumental in adding Latin American data and identity to AI.
The 2023-launched Latam-GPT works to tackle the linguistic biases of other generative AI models, which are primarily trained on English data. This project was announced at the Artificial Intelligence Action Summit in Paris, held in February 2025, and aims not just to rival existing AI models like ChatGPT, Copilot or Gemini, but also to provide the foundation for future regional generative AI applications.
According to Rodrigo Durán, Executive Director, CENIA, what sets Latam-GPT apart from its international peers is that this model has been trained on Latin American data, which was not previously available online and was therefore excluded from other such platforms. With region-specific data being fed to the model, Durán believes that this will result in more accurate and efficient responses and performance with respect to the LATAM and Caribbean regions.
The model has been trained on data privately sourced through strategic partnerships across the region, and also using synthetic data from areas which were previously earmarked as underrepresented. In order to develop this language model, data worth more than eight terabytes was collected, which equals millions of books, said Gabriela Arriagada, CENIA researcher and Latam-GPT’s Ethics Team Head.
Arriagada went on to explain that Latin American culture has been incorporated into this indigenously developed AI model to depict the various existing cultural realities, address the lacunae and shortcomings that persist in other such platforms, and build knowledge systems that ensure accurate and nuanced representation in the AI technology arena. However, there are some reservations about Latam-GPT’s ability to compete with larger corporations in the international market.
Latam-GPT is being hailed as a milestone in AI development, particularly because it contains data that captures each of the region’s eight countries’ particularities. Presently, it will operate in Spanish and Portuguese and will incorporate other indigenous languages down the line.
Durán explained that this development proves that the LATAM region can collaborate and develop such AI infrastructure. Additionally, creating such technologies also facilitates better regulations, as one cannot regulate something without completely understanding it.
AI race is dominating global politics and global markets. According to a recent study conducted by Oxford University, the US, China and Europe are leading the race with over half the world’s most powerful data centres, while South America and Africa have almost no AI hubs, and lag far behind in their technological capabilities.
In recent years, Chile has been rushing to construct new data centres and hire new experts in an effort to increase its share of the AI boom. Chilean President Boric said in June 2025 that a country that does not invest in AI risks falling behind in tomorrow’s global landscape.
Latam-GPT was created with just $550,000 from CENIA’s budget and the Development Bank of Latin America (CAF). The team’s initial version, which was developed using Amazon Web Services (AWS) cloud, will be accessible by the end of this month. Starting in the first semester of 2026, the latter versions will be trained on a $4.5 million supercomputer at the northern Chilean University of Tarapacá.
