Tucano 2 Cool: Better Open Source LLMs for Portuguese

Researchers have released Tucano 2, an open-source suite of Portuguese language models ranging from 0.5-3.7 billion parameters, featuring enhanced datasets and training recipes. The models achieve state-of-the-art performance on Portuguese benchmarks and include capabilities for coding, tool use, and chain-of-thought reasoning.

• →Tucano 2 provides fully open-source large language models specifically designed for Portuguese language processing.
• →The suite includes three variants (Base, Instruct, and Think) with parameters ranging from 0.5 to 3.7 billion.
• →New datasets include GigaVerbo-v2 Synth for synthetic data and specialized post-training datasets for advanced capabilities.
• →All training recipes, logs, and source code are openly released for reproducibility and community extension.
• →The models achieve state-of-the-art performance on several Portuguese-language modeling benchmarks.