Chile will launch Latam-GPT on Tuesday, an open-source artificial intelligence model designed to counter bias in largely US-centric AI systems.
Developed by the Chilean National Center for Artificial Intelligence, Latam-GPT is trained on millions of data points from Latin America to better reflect the region’s cultural diversity. Chilean Science Minister Aldo Valle said the project aims to address “prejudices” and avoid portrayals of Latin America as culturally uniform.
CENIA Director Alvaro Soto said Latin American data make up only a small share of existing global AI models, resulting in stereotypical depictions of people in the region.
Unlike closed systems such as ChatGPT or Google’s Gemini, Latam-GPT is open-source, allowing developers to customize it for local needs. The project drew contributions from universities, libraries, governments, and civil society groups across Latin America.
Latam-GPT was trained on more than eight terabytes of data and developed for about $550,000, funded mainly by the Development Bank of Latin America. It is currently trained in Spanish and Portuguese, with plans to include Indigenous languages.
While it is not expected to rival major global AI models, developers and users say Latam-GPT’s strength lies in its ability to understand local slang, idioms, and cultural context, making it useful for region-specific applications such as customer service and public services.