Chile has unveiled Latam-GPT, the first open-source artificial intelligence language model designed specifically for Latin America, in a strategic bid to reduce the region's dependence on US and Chinese technology and address cultural biases in existing AI systems, AFP reported.

The project, co-ordinated by Chile's National Center of Artificial Intelligence (CENIA) and supported by more than 60 institutions and nearly 200 specialists across eight countries, represents what organizers describe as a shift from Latin America being merely a consumer of AI foundation models to actively building and shaping them.

The model, which contains 70bn parameters and was developed on Meta's Llama 3.1 architecture, was trained primarily in Spanish and Portuguese, with plans to eventually incorporate indigenous languages. Unlike commercial chatbots such as ChatGPT or Google Gemini, Latam-GPT is positioned as foundational infrastructure: a public good that governments, universities, start-ups and companies can adapt and build upon.

"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 at the launch event on national television on February 10. He cited the disparity in available information between European historical events and Latin American milestones as evidence of the representation gap the project aims to address.

The initiative tackles what developers see as structural weaknesses in existing large language models, which are trained predominantly on English-language data and fail to adequately capture the region's languages, cultural nuances and real-world contexts. This can result in weaker performance on region-specific knowledge, higher bias risk and reduced policy control when deploying AI at national scale.

Development of Latam-GPT involved processing more than 8 terabytes of data—equivalent to millions of books—including private sources and synthetic datasets created to fill gaps in available information. The work was funded with $550,000 from CENIA's budget and the Development Bank of Latin America (CAF).

Initial training used Amazon Web Services cloud infrastructure, but subsequent versions will run on a $4.5mn supercomputer being installed at the University of Tarapacá in northern Chile, expected to be operational in the first half of 2026.

Álvaro Soto, CENIA's director, said models built elsewhere contain relatively little information about Latin America, limiting their effectiveness for regional applications. "The aim is to provide more accurate and culturally relevant outputs that reflect the realities of the region," he said.

Luis Chiruzzo, an engineering professor at the University of the Republic in Uruguay who participated in the project, described it as "a very important milestone for Latin America" that positions the region to engage with AI technologies on its own terms.

The open-access model is designed for use in text-intensive workflows common across public administration and services, including document drafting and summarization, translation, knowledge retrieval and customer support. Among early commercial users is Chilean firm Digevo, which is building AI-powered customer service tools for airlines and retailers that can understand regional dialects and local expressions.

Soto said the model could enable digital solutions tailored to local needs, such as tools for hospitals facing logistical challenges or resource allocation issues.

However, Alejandro Barros, an academic at the University of Chile, cautioned that Latam-GPT cannot realistically compete with large global AI models due to vast differences in economic resources and infrastructure.

The project's reliance on commercial cloud infrastructure for initial development also raises questions about long-term independence and sustainability, despite the model's framing as a sovereignty-building initiative.

Latam-GPT joins other regional AI efforts including SEA-LION in Southeast Asia and UlizaLlama in Africa, as developing regions seek to build AI capabilities that reflect their own cultural and linguistic contexts.