Portuguese Language Models

Latest papers

18 papers1 paper in the last four weeks, against 1 the four weeks before. 0.0% of all new papers.

Manacá-1B-Instruct-2: Reproducible Safety Alignment for a Native Brazilian-Portuguese LLM. Negative DPO Results, a Controlled Composition Scan, and a Silent Failure Mode of the LLM Judge
Oct 14, 2026Bruno Leonardo Santos Menezes, Carlos Leonardo Souza Cardoso, Fabio André Machado PortoPortuguese Language Models LLM Safety Alignment
National Laboratory for Scientific Computing (LNCC), Institute of Artificial Intelligence, Petrópolis, Brazil
Manacá-1B: An Open, Reproducible Brazilian-Portuguese Language Model and a Tokenizer-Aware, Paired Evaluation
Aug 31, 2026Bruno Leonardo Santos Menezes, Carlos Leonardo Souza Cardoso, Fabio Andre Machado PortoLanguage Model Pretraining Portuguese Language Models
Laboratório Nacional de Computação Científica (LNCC), Petrópolis, Brazil
Language Models for Portuguese: A Systematic Mapping Study
Aug 3, 2026Jhessica Silva, Carlos Caetano, Helena Maia +3Language Model Pretraining Portuguese Language Models
Instituto de Computação, Universidade Estadual de Campinas (UNICAMP), Campinas, 13083-852, São Paulo, Brazil
MTEB-BR: A Text Embedding Benchmark for Brazilian Portuguese
Jul 6, 2026Tardelli Ronan Coelho StekelText Embedding Evaluation Text Embedding Models
Federal Institute of S˜ao Paulo (IFSP) S˜ao Paulo, Brazil
Beyond Multilingual Averages: MTEB-PT, a Benchmark for Portuguese Sentence Encoders
Jul 5, 2026Lucas Hideki Takeuchi Okamura, Alexandre Alcoforado, Anna Helena Reali CostaSentence Embeddings Text Embedding Evaluation
Escola Politécnica, Universidade de São Paulo, São Paulo, Brazil
BERTomelo: Your Portuguese Encoder Best Friend
Jun 27, 2026Rennê Ruan Alves Oliveira, Gustavo Cordeiro Galvão Van Erven, Luís Paulo Faina GarciaLanguage Model Pretraining Portuguese Language Models
Department of Computer Science University of Brasília Brasília - DF, Brazil
moBERTo: A Modern Encoder for Portuguese via Continued Pretraining of ModernBERT
Jun 21, 2026Thiago Laitz, Thales Sales Almeida, João Guilherme Alves Santos +1Language Model Pretraining Portuguese Language Models
UNICAMP, Campinas, Brazil · Tropic AI · Maritaca AI
AMALIA-VL: A Native European Portuguese Open-Source Vision and Language Model
Jun 17, 2026Diogo Glória-Silva, João Cardeira, Manuel Letras da Luz +8Vision-Language Models Large Vision-Language Models
1NOVA School of Science and Technology · 2NOVA LINCS
P3B3: A Multi-Turn Conversational Benchmark for Measuring European and Brazilian Portuguese Variety Bias in LLMs
Jun 15, 2026Rafael Ferreira, Inês Vieira, Inês Calvo +6Multilingual Language Model Evaluation Social Bias in Language Models
NOVA University of Lisbon, Portugal · NOVA LINCS
PortBERT: Navigating the Depths of Portuguese Language Models
Jun 1, 2026Raphael Scheible-Schmitt, Henry He, Armando B. MendesLanguage Model Pretraining Portuguese Language Models
School of Computation, Information and Technology, Technical University of Munich, Munich, Germany · IS2E - Intelligent Systems, Science and Engineering, LIACC polo on Azores University, Ponta Delgada, Portugal
NorBERTo: A ModernBERT Model Trained for Portuguese with 331 Billion Tokens Corpus
Apr 30, 2026Enzo S. N. Silva, Pablo B. Costa, Raphael C. Vlasman +8Language Model Pretraining Portuguese Language Models
Itaú Unibanco, São Paulo, SP, Brasil · ICTi, São Paulo, SP, Brasil
ALBA: A European Portuguese Benchmark for Evaluating Language and Linguistic Dimensions in Generative LLMs
Mar 27, 2026Inês Vieira, Inês Calvo, Iago Paulo +6LLM Evaluation Language Model Generation Evaluation
1NOVA University of Lisbon, Portugal · 2NOVA LINCS
AMALIA Technical Report: A Fully Open Source Large Language Model for European Portuguese
Mar 27, 2026Afonso Simplício, Gonçalo Vinagre, Miguel Moura Ramos +19Portuguese Language Models LLM Training
NOVA School of Science and Technology · NOVA LINCS · Instituto de Telecomunicações +2
CAPITU: A Benchmark for Evaluating Instruction-Following in Brazilian Portuguese with Literary Context
Mar 23, 2026Giovana Kerche Bonás, Roseval Malaquias Junior, Marcos Piau +6Instruction Following Multilingual Language Model Evaluation
Maritaca AI · Jusbrasil
Efficient Fine-Tuning Methods for Portuguese Question Answering: A Comparative Study of PEFT on BERTimbau and Exploratory Evaluation of Generative LLMs
Mar 22, 2026Mariela M. Nina, Caio Veloso Costa, Lilian Berton +1Parameter-Efficient Fine-Tuning Question Answering
Institute of Science and Technology, Federal University of São Paulo (UNIFESP), São José dos Campos, SP, Brazil
CLARIN-PT-LDB: An Open LLM Leaderboard for Portuguese to assess Language, Culture and Civility
Mar 13, 2026João Silva, Luís Gomes, António BrancoLLM Evaluation Portuguese Language Models
University of Lisbon NLX—Grupo de Fala e Linguagem Natural, Departmento de Informática Faculdade de Ciências, Campo Grande, 1749-016 Lisboa, Portugal
Sabiá-4 Technical Report
Mar 10, 2026Thiago Laitz, Thales Sales Almeida, Hugo Abonizio +6Portuguese Language Models Language Modeling
Maritaca AI · Jusbrasil
Tucano 2 Cool: Better Open Source LLMs for Portuguese
Mar 3, 2026Nicholas Kluge Corrêa, Aniket Sen, Shiza Fatimah +4Portuguese Language Models Language Modeling
Bonn-Aachen International Center for Information Technology (b-it) / CAISA Lab · Lamarr Institute for Machine Learning and Artificial Intelligence · Center for Science and Thought +2

Oct 14, 2026J/K move · Enter open · S save

Manacá-1B-Instruct-2: Reproducible Safety Alignment for a Native Brazilian-Portuguese LLM. Negative DPO Results, a Controlled Composition Scan, and a Silent Failure Mode of the LLM Judge

Bruno Leonardo Santos Menezes, Carlos Leonardo Souza Cardoso, Fabio André Machado Porto

National Laboratory for Scientific Computing (LNCC), Institute of Artificial Intelligence, Petrópolis, Brazil

Portuguese Language Models LLM Safety Alignment Language Modeling Language Model Safety Evaluation Over-Refusal Direct Preference Optimization+1

Open paper AskPDF arXiv

Brazilian Portuguese remains underserved by open-source language models, and the few that are safety-aligned rarely document what did not work or validate comparisons with statistical significance. We present Manacá-1B-Instruct-2, the second-generation post-training alignment of Manacá-1B (1.72B parameters, trained from scratch for PT-BR), with an SFT corpus of 291,223 examples. We document three DPO attempts, all negative, including a case where the training reward margin reaches 1.0 while free generation does not change at all, and show that safety came from a dedicated safety-SFT (LLM-jp AnswerCarefully style), refined over three iterations until closing the safety-helpfulness frontier. A controlled scan of the fraction of general data in the safety-SFT mix (10% to 60%) reveals a non-monotonic over-refusal curve with a knee point at 40-50%. The final model is compared against four external PT-BR models on the same items, with Wilson confidence intervals and paired McNemar tests with Holm-Bonferroni correction: the lower bound of its safety interval (94.1%) lies above the upper bound of all four baselines. We also report a methodological finding for LLM-judge evaluation: on two of six safety items in MT-Bench-PT, the judge categorically refuses to assign a score regardless of the evaluated response, silently reducing the effective N of that category in every prior report of the project. We complement this with an n-gram decontamination check and a Chinchilla-anchored compute-budget table that attributes the gap in general capability to pre-training scale rather than alignment methodology. All code, raw data, logs, and scripts that produce every reported number are publicly released.