In Chile, we are convinced that innovation and technology are the key tools to build a more inclusive, sustainable and competitive future,” Chilean President Gabriel Boric said when announcing the artificial intelligence project. file Photo by Elvis Gonzalez/EPA
Latam GPT, the first open-source artificial intelligence model developed in Latin America, has begun training and is processing more than 1 billion documents ahead of its planned launch at the end of the year.
The project is a large language model coordinated in Chile by the National Center for Artificial Intelligence , or Cenia, in partnership with Brazil. Financial support comes from the Development Bank of Latin America and the Caribbean.
It will differ from models such as ChatGPT, Claude.ai and Gemini because it was not built on English. The open-source model is designed to reflect Latin America’s culture, language and history, and aims to preserve Indigenous languages such as Mapudungun and Rapa Nui, as well as regional dialects.
“The context it provides should be stronger because it has been trained on more data from the region. If I need to ask something about Latin America and the Caribbean, Latam GPT should perform better than other models because it will have more data on that,” Rodrigo Durán, general manager of Cenia, a private nonprofit founded by Chilean universities, told UPI.
Durán stressed the importance of preserving local languages and dialects.
“Language models are increasingly a source of access to knowledge and research. So, if information specific to those cultures is not available to these models — whether through a project like Latam GPT or another — it is unlikely that culture will be preserved and shared. That is why this work is essential,” he said.
He added that countries that do not invest in artificial intelligence risk falling behind on the global stage.
“In Chile, we are convinced that innovation and technology are the key tools to build a more inclusive, sustainable and competitive future,” Chilean President Gabriel Boric said when announcing the project. He added, “The digital future must also speak in our language, with our voices and for our people.”
The AI is expected to handle 50 trillion parameters, comparable to OpenAI’s ChatGPT 3.5. It is now in the third of eight stages, which involve gathering data from libraries, government agencies, universities and institutions with information in Spanish, Portuguese and English.
“Cenia has already collected nearly 500 gigabytes of data through partnerships in Spanish and Portuguese. Our mission is to process a total of 20.5 terabytes of public data in English by the end of the project,” said Mauricio Leiva, a computer engineer and project manager for Latam GPT.
Information has been compiled from across the region, including web data, blogs, news sites, academic articles and educational resources in fields such as arts, science, sports, education, medicine and politics, among others.
The launch of this technological milestone is scheduled for the fourth quarter of 2025.
“A beta version of the model will be released in October for academic review and fine-tuning. In December 2025, the first official version will be launched and made available to researchers, students and the general public,” Leiva said.
By 2026, the team plans to develop an interface that allows easier interaction with users, “like ChatGPT,” he said.
Executives said one of the main challenges in developing the project has been securing the necessary computing power while working with limited financial resources.
On funding, Durán said their investment is far lower than that of major companies, such as Google or OpenAI.
“The training cost of Gemini Ultra, Google’s language model, was close to $200 million in computing alone, while ChatGPT-4 cost $171 million. Latam GPT will be about $3.5 million — roughly 80 times fewer resources for the entire project. The challenge has been how to build a competitive model with the constraints we face,” he said.
