Summary: Chilean President Gabriel Boric launched Latam-GPT on February 10, 2026 as the first large language model built inside Latin America. CENIA led a coalition spanning 15 countries, and the 70-billion-parameter model based on Meta’s Llama 3.1 was released under an open license. Its reported budget was about $550,000, far below frontier-lab spending, and training used 2.6 million regional documents.
February 10, 2026 marked a first for the region: Chilean President Gabriel Boric took the wraps off Latam-GPT, the first large language model to be built inside Latin America. Steering the effort is Chile’s National Center for Artificial Intelligence (CENIA) under director Álvaro Soto; the coalition behind it spans 15 countries, more than 200 collaborators, and 33 institutional alliances, drawing in universities, foundations, libraries, and government bodies from Argentina, Brazil, Colombia, Ecuador, Mexico, Peru, Uruguay, and beyond. The money came from two directions: CENIA’s own resources and the Development Bank of Latin America (CAF).
For developers, the positioning is the real story, not the launch ceremony. This is not a chatbot built for end consumers; it is open infrastructure — a 70-billion-parameter model built on Meta’s Llama 3.1, shipped under an open license, free for companies and public institutions to download and customize, and hosted on the official site at latamgpt.org.
Training on 2.6 Million Regional Documents
What separates Latam-GPT from the pack is the data. Training consumed more than 8 terabytes of regional material — 2.6 million documents gathered from 20 Latin American countries plus Spain — with Spanish and Portuguese supported today and Indigenous Latin American languages on the way. Behind the effort sits a data gap long left unaddressed: Spanish and Portuguese together reportedly account for a mere 2–3 percent of the training data behind existing large models, which is why outputs naturally skew toward a US and Global North worldview.
Soto is blunt about it: “No matter how powerful the large models are, they cannot cover all aspects relevant to our reality.” The example he reaches for is cultural blindness — by default, models orbit figures like George Washington and US-centric framings instead of the historical figures, idioms, and local context a Latin American user actually needs.
The Sovereignty Case: A Seat at the Table, Not a Spot on the Menu
At the launch, Boric pitched the project in terms of sovereignty, arguing that Latam-GPT positions the region as “an active and sovereign player in the economy of the future.” The line that traveled furthest: “We’re at the table — we’re not on the menu.” Tech Policy Press also cited his description of the project as “defending our identity and our right to exist.”
In 2026, that argument lands hard. Frontier capability and pricing are concentrated in a handful of US and Chinese labs, which leaves a region seeking autonomy two realistic levers: apply regulatory pressure, or build the model itself. Latam-GPT went with building — and paired it with open source — so any government, media outlet, or startup can download the weights and fine-tune them instead of depending on one vendor’s API.
What a $550,000 Budget Buys
The balance sheet is less romantic. Reported total investment sits near $550,000 — roughly $300,000 from CENIA and $250,000 from CAF — leaving the project more than three orders of magnitude behind the multi-billion-dollar training budgets of US and Chinese labs. AWS carried the training for the first version; future runs are headed to a supercomputer at the University of Tarapacá in Arica, northern Chile, where a $10 million investment has produced the region’s first computing center capable of training large models on domestic soil.
Applications are already taking concrete form. Soto points to hospital logistics as one candidate workload, and entrepreneur Roberto Musso’s company Digevo plans customer-service tooling that picks up on regional “slang, idioms, and even speech rate” — demand that global models leave poorly served even though the market plainly exists.
Zoomed Out: The Global Open-Weights Picture
Skeptics see two hazards. One is resources: OpenAI, Google, and Anthropic command such overwhelming means that cheaply fine-tuning existing models on Spanish and Portuguese data would be enough to erase Latam-GPT’s regional edge. The other is politics: instability in the region could threaten the project’s long-term continuity, and the failed 2009 UNASUR fiber-optic project is the cautionary tale locals always bring up. Tech Policy Press’s verdict: Latam-GPT’s real value lies in building regional capacity and expertise, not in going head-to-head with global giants.
Zoomed out against the global open-weights landscape, that reading gains another dimension. The US and China keep shipping open-weight models on the performance track — MiniMax’s M2.5 open-weights coding model is a case in point — while Latam-GPT runs the data-sovereignty track. For product teams the takeaway is practical: if you serve Spanish- or Portuguese-speaking markets, there is now a fully customizable base model with no API bill attached. Its value isn’t beating GPT at benchmarks; it’s that nobody can take it away from you.