Latam-GPT
Open-weight 70B LLM built by 60+ Latin American institutions
Overview
Latam-GPT is an open-source LLM coordinated by Chile's CENIA with 60+ institutions across 15 Latin American and Caribbean countries. Built on Llama 3.1 70B and fine-tuned on regional Spanish and Portuguese text, its weights are public on Hugging Face for researchers and developers.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
What you can produce with Latam-GPT
- Open-weight 70B model (Llama 3.1 base) on Hugging Face
- Spanish and Portuguese language fine-tuning
- Data pipeline tooling (toxicity classification, deduplication, anonymization) on GitHub
- Quantized GGUF community builds for local inference
- Coordinated by CENIA with 60+ regional institutions
ASEAN Perspective
Latam-GPT in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Latam-GPT is a rare case of a government-and-academia-led open model that actually shipped weights instead of just holding a press conference — Llama-3.1-70B-LatamGPT-SFT-1.0 is live on Hugging Face, coordinated by Chile's CENIA with universities across more than a dozen countries pooling data and compute, including a new domestically built supercomputing center in Arica. The training corpus runs to roughly 70 billion words of Spanish and Portuguese text, aimed at cultural and historical context that foreign models tend to miss.
It's explicitly not a ChatGPT competitor: there's no consumer app, no hosted API for casual use, and a full technical report with benchmark numbers hadn't shipped as of this writing. As a Llama 3.1 derivative, expect roughly Llama-3.1-class general capability with better regional grounding, not frontier performance. The real audience is regional developers, governments, and researchers who want a sovereign base model to build on, not end users looking for a chatbot.
What people say
Latam-GPT launched publicly in February 2026 with Chilean President Gabriel Boric at the unveiling, positioned as the first LLM built end-to-end in Latin America rather than adapted from a US or Chinese base by a single company. The coordination effort is unusually large for an open-source AI project: more than 60 institutions across roughly 15 countries contributed data, evaluation work, and compute under CENIA's coordination.
Technically, it's a continued-pretraining and supervised-fine-tuning pass on Meta's Llama 3.1 70B, using a training corpus reported at roughly 70 billion words of Spanish- and Portuguese-language text plus some Indigenous-language material, explicitly aimed at reducing the US-centric cultural bias that general-purpose Western models carry into Latin American use cases. The University of Tarapacá in Arica built a $10M supercomputing facility specifically to support the training runs domestically, rather than relying on rented cloud GPU capacity abroad.
Weights for Llama-3.1-70B-LatamGPT-SFT-1.0 are public on Hugging Face under the latam-gpt organization, with quantized GGUF community conversions already circulating for local inference. The GitHub organization hosts the surrounding tooling rather than the model itself — data filtering, toxicity classification, language identification, anonymization, and deduplication scripts used to build the training corpus — a useful signal the project treats data-pipeline work as a first-class deliverable, not an afterthought.
Independent commentary from Brookings and Tech Policy Press frames the project as a genuine AI-sovereignty move but flags the gap between regional ambition and what a 70B Llama derivative can realistically compete with against frontier commercial models. A full technical report with benchmark scores had not been published as of these searches, so claims about how well it handles Latin American cultural and historical context specifically aren't yet independently measurable. It is explicitly not built as a public-facing chatbot — there's no official hosted consumer app — so most people will encounter it only through downstream tools built on the released weights.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Latam-GPT's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Latam-GPT unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Latam-GPT directly →
Spotted something out of date? Suggest an update →
Latam-GPT in the news
Alternatives to Latam-GPT
More in LLMs & Chat
