Sabiá-3 Technical Report

2024/10/15 by Hugo Abonizio, Abonizio, Hugo, Thales Sales Almeida +10 · 3 citations
Engineering · Physics and Astronomy · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Fluid Dynamics Simulations and Interactions #Historical Astronomy and Related Studies

paper · pdf · doi:10.48550/arxiv.2410.12049

openalex publication_date 2024/10/15 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/03

Abstract

This report presents Sabiá-3, our new flagship language model, and Sabiazinho-3, a more cost-effective sibling. The models were trained on a large brazilian-centric corpus. Evaluations across diverse professional and academic benchmarks show a strong performance on Portuguese and Brazil-related tasks. Sabiá-3 shows large improvements in comparison to our previous best of model, Sabia-2 Medium, especially in reasoning-intensive tasks. Notably, Sabiá-3's average performance matches frontier LLMs, while it is offered at a three to four times lower cost per token, reinforcing the benefits of domain specialization.

Cited by

• PoETa v2: Toward More Robust Evaluation of Large Language Models in Portuguese
• Ticket-Bench: A Kickoff for Multilingual and Regionalized Agent Evaluation
• BRoverbs -- Measuring how much LLMs understand Portuguese proverbs

Related

• Sabiá-2: A New Generation of Portuguese Large Language Models
• Juru: Legal Brazilian Large Language Model from Reputable Sources
• Cabrita: closing the gap for foreign languages
• Amadeus-Verbo Technical Report: The powerful Qwen2.5 family models trained in Portuguese
• SLIM-RAFT: A Novel Fine-Tuning Approach to Improve Cross-Linguistic Performance for Mercosur Common Nomenclature