Sabiá-4 Technical Report

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Laitz, Thiago, Almeida, Thales Sales, Abonizio, Hugo, Junior, Roseval Malaquias, Bonás, Giovana Kerche, Piau, Marcos, Larcher, Celio, Pires, Ramon, Nogueira, Rodrigo
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911504208494592
author Laitz, Thiago
Almeida, Thales Sales
Abonizio, Hugo
Junior, Roseval Malaquias
Bonás, Giovana Kerche
Piau, Marcos
Larcher, Celio
Pires, Ramon
Nogueira, Rodrigo
author_facet Laitz, Thiago
Almeida, Thales Sales
Abonizio, Hugo
Junior, Roseval Malaquias
Bonás, Giovana Kerche
Piau, Marcos
Larcher, Celio
Pires, Ramon
Nogueira, Rodrigo
contents This technical report presents Sabiá-4 and Sabiazinho-4, a new generation of Portuguese language models with a focus on Brazilian Portuguese language. The models were developed through a four-stage training pipeline: continued pre-training on Portuguese and Brazilian legal corpora, long-context extension to 128K tokens, supervised fine-tuning on instruction data spanning chat, code, legal tasks, and function calling, and preference alignment. We evaluate the models on six benchmark categories: conversational capabilities in Brazilian Portuguese, knowledge of Brazilian legislation, long-context understanding, instruction following, standardized exams, and agentic capabilities including tool use and web navigation. Results show that Sabiá-4 and Sabiazinho-4 achieve a favorable cost-performance trade-off compared to other models, positioning them in the upper-left region of the pricing-accuracy chart. The models show improvements over previous generations in legal document drafting, multi-turn dialogue quality, and agentic task completion.
format Preprint
id arxiv_https___arxiv_org_abs_2603_10213
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Sabiá-4 Technical Report
Laitz, Thiago
Almeida, Thales Sales
Abonizio, Hugo
Junior, Roseval Malaquias
Bonás, Giovana Kerche
Piau, Marcos
Larcher, Celio
Pires, Ramon
Nogueira, Rodrigo
Computation and Language
This technical report presents Sabiá-4 and Sabiazinho-4, a new generation of Portuguese language models with a focus on Brazilian Portuguese language. The models were developed through a four-stage training pipeline: continued pre-training on Portuguese and Brazilian legal corpora, long-context extension to 128K tokens, supervised fine-tuning on instruction data spanning chat, code, legal tasks, and function calling, and preference alignment. We evaluate the models on six benchmark categories: conversational capabilities in Brazilian Portuguese, knowledge of Brazilian legislation, long-context understanding, instruction following, standardized exams, and agentic capabilities including tool use and web navigation. Results show that Sabiá-4 and Sabiazinho-4 achieve a favorable cost-performance trade-off compared to other models, positioning them in the upper-left region of the pricing-accuracy chart. The models show improvements over previous generations in legal document drafting, multi-turn dialogue quality, and agentic task completion.
title Sabiá-4 Technical Report
topic Computation and Language
url https://arxiv.org/abs/2603.10213