TranslateGemma - Google's Open-Source Series of Translation Models

What Is TranslateGemma

TranslateGemma is a new generation of open-source series of translation models from Google, built on Gemma 3 . Through a combination of supervised fine-tuning and reinforcement learning, the models support high-quality translation across 55 languages, significantly improving translation accuracy and efficiency. TranslateGemma offers models at three parameter scales—4B, 12B, and 27B—suited respectively to mobile devices, consumer laptops, and cloud high-performance computing environments. The models perform excellently in text translation and retain strong multimodal capabilities, able to translate text in images without specialized multimodal training.

TranslateGemma's Main Features

• Multilingual Translation: The models support high-quality translation across 55 languages, covering both high-resource languages (such as English, Chinese, and French) and low-resource languages (such as Icelandic and Swahili).
• High Efficiency: Through model optimization, the smaller 12B model outperforms the 27B baseline model, while the 4B model delivers strong translation capabilities on mobile devices.
• Multimodal Translation: The models retain the multimodal capabilities of Gemma 3 and can translate text in images without additional multimodal training.
• Broad Applicability: The models are suitable for a variety of deployment environments, including mobile devices, consumer laptops, and cloud high-performance computing platforms.

TranslateGemma's Technical Principles

• Supervised Fine-Tuning (SFT)：The models are fine-tuned with high-quality parallel data, including human-translated text and synthetic translation data generated by the Gemini model.In this way, the models significantly improve translation quality on low-resource languages while maintaining general capabilities.
• Reinforcement Learning (RL)：The models optimize translation quality using multiple reward models (such as MetricX-QE and AutoMQM).Through reinforcement learning, the models can generate more natural and accurate translations, further improving translation quality.

TranslateGemma's Project Address

• Project Official Website：https://blog.google/innovation-and-ai/technology/developers-tools/translategemma/
• HuggingFace Model Repository：https://huggingface.co/collections/google/translategemma
• arXiv Technical Paper：https://arxiv.org/pdf/2601.09012

TranslateGemma's Application Scenarios

• Personal Communication and Travel: The models help individuals overcome language barriers in instant messaging and travel, translating chat content in real time or translating menus, road signs, and more.
• Content Creation and Distribution: News media and the film and television industry can quickly translate content into multiple languages, expanding international influence and enhancing the global accessibility of content.
• Corporate Internationalization: Enterprises use translation tools to provide multilingual customer support, optimize marketing copy, improve the efficiency of cross-national collaboration, and support global development.
• Education: The models can assist students in learning foreign languages and help researchers quickly translate foreign-language literature, promoting language learning and academic exchange.
• Multimodal Applications: Automatically translate text in image and video content, such as translating road signs or video subtitles, improving the comprehensibility of multimodal content.