TranslateGemma: A new family of open translation models
Google introduces TranslateGemma, a new family of open translation models built on Gemma 3, available in 4B, 12B, and 27B parameter sizes. These models, which cover 55 languages, are designed for diverse deployment environments and offer high-fidelity translation quality with greater efficiency. They outperform larger models in some cases and are built using a specialized two-stage fine-tuning process. Researchers and developers can access these models through Kaggle, Hugging Face, and Vertex AI.