Latin America & Iberia

AI models, vendors, and hardware buying realities from Latin America & Iberia. Curated by myaihardware.com.

Models

Vendors

Languages

Why Latin America & Iberia matters for AI builders

Latin America offers AI builders direct access to specialized AI hardware like NVIDIA H100s through local cloud providers, bypassing US export delays for many workloads. The region's AI ecosystem leans heavily on open-weight models such as Llama and Mistral, reducing dependency on proprietary APIs. An underrated angle: LatAm's unique mix of high renewable energy capacity and relatively cheap land makes it a natural site for sustainable, low-cost inference clusters.

Market context

Brazil's AI strategy (Plano Brasileiro de IA / PBIA, 2024) directs ~R$23 billion to sovereign AI infrastructure, with Maritaca AI (Sabiá, MariTalk) emerging as the leading commercial Portuguese-first lab while federal universities ship open-weight finetunes (Bode, Cabra, Caramelo, Tucano, Quati) under MIT/Apache licenses. Portugal's open-LLM work is academia-driven through PORTULAN at INESC-ID Lisboa, which releases the Albertina family for both PT-PT and PT-BR variants under FCT (Fundação para a Ciência e a Tecnologia) funding. Spain's Barcelona Supercomputing Center (BSC) anchors EU-funded ALIA and Salamandra programs delivering 7B/40B multilingual checkpoints across Spanish, Catalan, Galician, Basque and Romani, with significant Spanish-government and EU-Horizon co-investment. Hardware reality for LATAM hobbyists is harsh — Argentine import duties exceed 50%, Brazilian RTX 4090s sell at ~2x US MSRP, and Mexican developers face USD-denominated cloud bills against weak local currency, pushing the region toward smaller 7B models, aggressive quantization, and shared-compute arrangements at university clusters.

Models (23 of 23)

Filter by vendor, license, or parameter range.

LicenseVendorParameter range

MariTalk 7B

Maritaca AI · Brazil

Restricted

Params: 7B
Context: 8,192
Min VRAM: 14 GB
GGUF: No

Strong Portuguese understanding, fine-tuned for Brazilian Portuguese

Sabiá 7B

Maritaca AI · Brazil

Restricted

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: No

Early Brazilian LLM, good for Portuguese tasks

Sabiá 3

Maritaca AI · Brazil

Restricted

Params: 3B
Context: 4,096
Min VRAM: 6 GB
GGUF: No

Lightweight, efficient for edge devices

Bode Alpaca PT

UFSCar / Federal University of São Carlos · Brazil

Open

Params: 7B
Context: 2,048
Min VRAM: 14 GB
GGUF: Yes

Open source, fine-tuned for Portuguese instructions

Cabra 7B

Federal University of Ceará (UFC) · Brazil

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Brazilian Portuguese focus, open source

Caramelo 7B

Federal University of Rio Grande do Sul (UFRGS) · Brazil

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Optimized for Brazilian Portuguese, open weights

Tucano 7B

Federal University of Amazonas (UFAM) · Brazil

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Regional focus on Amazonian Portuguese, open source

Lince Mistral 7B

Lince AI / University of São Paulo (USP) · Brazil

Open

Params: 7B
Context: 8,192
Min VRAM: 14 GB
GGUF: Yes

Multilingual (Portuguese, Spanish, English), Mistral architecture

Quati 7B

Federal University of Minas Gerais (UFMG) · Brazil

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Brazilian Portuguese focus, open source

JurisGPT BR

University of Brasília (UnB) · Brazil

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Specialized in Brazilian legal domain

Albertina PT-BR

University of São Paulo (USP) / IBM Research · Brazil

Open

Params: 7B
Context: 512
Min VRAM: 6 GB
GGUF: No

Strong for Portuguese NLP tasks (classification, NER)

Albertina PT-PT

University of Lisbon / INESC-ID · Portugal

Open

Params: 7B
Context: 512
Min VRAM: 6 GB
GGUF: No

European Portuguese focus, strong for NLP tasks

Albertina PT-PT 900M

University of Lisbon / INESC-ID · Portugal

Open

Params: 900M
Context: 512
Min VRAM: 2 GB
GGUF: No

Lightweight, efficient for European Portuguese

Glória Clinical

Hospital Israelita Albert Einstein / USP · Brazil

Restricted

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: No

Specialized in clinical Portuguese, medical domain

Alia 7B

Alia AI (Spain) · Spain

Open

Params: 7B
Context: 8,192
Min VRAM: 14 GB
GGUF: Yes

Multilingual for Spain (Spanish, Catalan, Galician, Basque), open source

Alia 40B

Alia AI (Spain) · Spain

Open

Params: 40B
Context: 16,384
Min VRAM: 80 GB
GGUF: No

Large model, strong multilingual performance for Spain

Salamandra 7B

BSC (Barcelona Supercomputing Center) · Spain

Open

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: Yes

Government-backed, strong for Spanish and Catalan

Salamandra 40B

BSC (Barcelona Supercomputing Center) · Spain

Open

Params: 40B
Context: 16,384
Min VRAM: 80 GB
GGUF: No

Large-scale, government-backed, strong for Iberian languages

Aya 23 8B

Cohere For AI · multinational

Open

Params: 8B
Context: 8,192
Min VRAM: 16 GB
GGUF: Yes

Strong multilingual performance, especially in low-resource languages; good instruction following

Aya 23 35B

Cohere For AI · multinational

Open

Params: 35B
Context: 8,192
Min VRAM: 70 GB
GGUF: Yes

Higher accuracy on complex tasks; better reasoning than 8B variant

Llama2 Pampero

Argentina (fine-tune) · Argentina

Restricted

Params: 7B
Context: 4,096
Min VRAM: 14 GB
GGUF: No

Optimized for Argentinian Spanish; captures local idioms and slang

RedRomaNi

Barcelona Supercomputing Center (Spain) · Spain

Open

Params: 7B
Context: 8,192
Min VRAM: 14 GB
GGUF: No

Supports Romani language; preserves cultural heritage

Lince Spanish Mistral

Spain (fine-tune) · Spain

Open

Params: 7B
Context: 8,192
Min VRAM: 14 GB
GGUF: Yes

High-quality Spanish language generation; good for conversational AI

Notable vendors

Maritaca AIBSC (Barcelona Supercomputing Center)PORTULAN / INESC-IDCohere For AIBrazilian federal universities (UFC, UFRGS, UFMG, UnB, UFAM, USP)

Run benchmarks

llama.cpp, MLPerf, and inference leaderboards.

VRAM calculator

Will this model fit on your GPU?

Hardware guides

Local LLM setup, homelab, mini-PC builds.

Methodology

How we benchmark, verify ASINs, and source data.

Beginner's guide

New to local AI? Start here.

Sourcing & commerce disclosure

Model metadata on this page is compiled from each vendor's public HuggingFace repo, model card, or research paper at time of indexing. Performance figures (KMMLU, MMLU, HAERAE, AraBench, IndicGEM, etc.) link directly to source benchmarks where available; missing or partial values are shown as ", ".

Outbound hardware links (GPUs, mini-PCs, accelerators) are Amazon Associate links using store ID fredoline-20. As an Amazon Associate we earn from qualifying purchases. Prices and availability shown elsewhere on the site are snapshots, not live quotes — verify on Amazon before purchasing. See About / disclosures for the full policy.