In a significant step toward AI sovereignty, a coalition of Latin American nations is set to launch its own open-source large language model, known as Latam-GPT, by September 2025. This initiative is headed by Chile’s National Centre for Artificial Intelligence (CENIA), bringing together more than 30 institutions across 12 countries.
Unlike other mainstream models, such as ChatGPT or Gemini, Latam-GPT is explicitly designed to reflect Latin America’s cultural, social, and linguistic diversity. Development began in early 2023 using Meta’s LLaMA 3 architecture, and the model is being trained on regionally sourced data across a decentralised cloud network, with infrastructure support from Chile’s University of Tarapacá and Amazon Web Services.
A key feature distinguishing Latam-GPT is its inclusion of Indigenous languages. The first integration is a Rapa Nui translator, which is already in progress, and further efforts are planned to include a broader spectrum of native dialects. This move aligns with the coalition’s mission to democratise access to AI in areas such as education, healthcare, and public services.
Though no dedicated budget has yet been confirmed and finalised, the regional development bank CAF and AWS have extended early support. The developers emphasise that Latam-GPT is not intended for commercial rivalry but rather serves as a foundational model for Latin America-specific applications. “It could be a democratizing element for AI,” said Chile’s Science Minister Aisén Etcheverry.
The full release is expected in September 2025 with open-access deployment and multi-institution testing underway. Project leaders hope the launch will attract additional funding and partnerships to accelerate the integration of indigenous communities and refine the model’s performance.
Latam-GPT marks a significant move toward tech independence, highlighting a global shift in which regions seek AI tools that reflect local priorities, languages, and ethics, rather than relying solely on Western-centric platforms.