Tired of ChatGPT, Latin America Develops Its Own AI

A New Era of AI in Latin America

In May, Juan Palma, a graphic designer based in Santiago, Chile, asked ChatGPT for directions to a nearby subway station. The response he received was not only incorrect but also sent him in the opposite direction. This experience highlighted a growing concern: while large language models like ChatGPT and Meta’s Llama are widely used, their accuracy and relevance in non-English contexts remain limited.

Addressing the Gaps

To tackle these limitations, over 30 institutions across Latin America have been working on a project called Latam-GPT. This open-source large language model is designed specifically for the region, incorporating local languages, dialects, and cultural nuances. It will be available to the public in September and is expected to provide more accurate and relevant responses for Latin American users.

Héctor Bravo, lead of disruptive technologies at Sonda, a Chilean IT firm, emphasized that the project aims to "build AI in Latin America, for Latin Americans." This means redefining success metrics beyond just accuracy or speed, focusing instead on cultural representation, social impact, and accessibility.

Deep Multilingualism and Cultural Inclusion

Latam-GPT is being developed with deep multilingualism in mind. It includes Indigenous languages such as Nahuatl, Quechua, and Mapudungun, as well as regional dialects from the Caribbean. This approach ensures that the model can understand and respond to the diverse linguistic landscape of Latin America.

Other regions have also taken steps toward localized AI. Southeast Asia has Sea-Lion, a family of open-source LLMs trained in multiple regional languages. In Africa, UlizaLlama supports several languages, including Xhosa and Zulu. India's BharatGPT supports over 14 regional languages, with the government also developing its own LLM.

Latin America's Growing AI Landscape

Latin America has been slower to adopt AI compared to other regions, but it is beginning to catch up. According to the Atlas of Artificial Intelligence for Latin America and the Caribbean, a 2025 report by the United Nations Development Programme, Chile leads in terms of regulation and institutional development.

Chile’s National Center for Artificial Intelligence (CENIA) was founded in 2021, and the idea for Latam-GPT emerged shortly after. Alvaro Soto, head of CENIA, explained that the initiative required broad collaboration from various stakeholders. His team began gathering databases and recruiting universities, government offices, and civil organizations.

Last year, CENIA signed 33 strategic alliances for Latam-GPT across Latin America, the U.S., and Spain, assembling 50 billion parameters — equivalent to ChatGPT 3.5.

Training with Local Data

Unlike global models that often rely on data from Spain or English translations, Latam-GPT is being trained with data from schools, businesses, libraries, and historical texts. This helps the model better understand the contexts and needs of Latin American users.

Omar Florez, technical lead for pre-training, explained that this approach ensures the model can grasp cultural and linguistic nuances that may be missed by other models.

Increasing Demand for Generative AI

There is a growing demand for generative AI platforms in the region. Brazil has the highest number of ChatGPT users after the U.S. and India, according to DemandSage. Llama downloads have also surged in Latin America, with teachers, students, and business owners using them for various purposes.

Even government offices use these tools to streamline processes. For example, courts in Buenos Aires use ChatGPT to draft legal decisions.

However, Latam-GPT faces challenges. It will be text-only for the foreseeable future and may lag behind in general knowledge and questions unrelated to Latin America. Carlos Honorato, CEO of Orión, noted that the project requires ultra-high-capacity infrastructure, specialized talent, and relevant datasets — areas where gaps still exist in the region.

Environmental and Legal Considerations

Environmental concerns also loom large. Large language models consume significant amounts of energy and water, prompting pushback against data centers in many countries, including Chile. The computing infrastructure for Latam-GPT is housed at the University of Tarapacá, a region suffering from decades of drought.

Despite this, the team at CENIA uses a flexible and scalable cloud-based infrastructure that optimizes resources and reduces energy use. They also utilize solar energy to minimize environmental impact.

Legal analysts are also concerned about the patchwork of data privacy legislation in Latin America. While Brazil has robust laws, neighboring Bolivia lacks comprehensive personal data protection regulations. These issues could lead to litigation and sanctions if not addressed properly.

Ensuring Inclusivity and Representation

Experts worry about whether Latam-GPT will accurately represent minorities and ensure access for historically marginalized groups. Varinka Farren, CEO of Hub APTA, emphasized the need for Indigenous peoples, migrant communities, and other groups to participate in the model’s validation.

Rodrigo Durán, CENIA’s general manager, stated that ensuring inclusivity is one of the project’s goals. While initial testing has been encouraging, it will likely take at least a decade to achieve full representation.

For Durán, the greatest contribution of Latam-GPT will be demonstrating that Latin America and the Caribbean have the capabilities and talent to carry out such an ambitious project.