Google on January 15, announced TranslateGemma, a new suite of open translation models built on Gemma 3, to support text translation across 55 languages.
The models are available in 4B, 12B, and 27B parameter sizes and are intended for use across mobile, local, and cloud environments, the company said. According to Google, TranslateGemma offers higher translation quality with fewer parameters by distilling capabilities from its larger Gemini models.
“By distilling the knowledge of our most advanced large models into compact, high-performance open models, we have created a suite where efficiency doesn’t require a compromise on quality,” the company said in its announcement.
TranslateGemma models are available through multiple channels, including Kaggle, Hugging Face, Vertex AI, and Google’s Gemma Cookbook.
The models were trained and evaluated across 55 language pairs, covering high-, mid-, and low-resource languages. Google said TranslateGemma reduced translation error rates across all tested languages compared to the baseline Gemma models. In addition, the company trained the system on nearly 500 more language pairs to allow researchers to fine-tune models for specific use cases.
Google said internal tests showed the 12B TranslateGemma model outperformed the larger Gemma 3 27B baseline on the WMT24++ benchmark using the MetricX framework. The 4B model delivered performance comparable to the 12B baseline, making it suitable for on-device inference.
Meanwhile, OpenAI has also introduced ChatGPT Translate, a web-based translation tool. The service is positioned as an alternative to Google Translate and uses ChatGPT models to translate text across more than 50 languages. It focuses on text input with automatic language detection and can be used without an account, though signing in enables additional features.
The tool uses a dual-panel layout, with one section for input text and another for the translated output. Users can select source and target languages from dropdown menus. ChatGPT Translate also allows users to adjust the tone of translations, such as making them more formal, simplified, or suited for academic use.
ALSO READ: Gemini Introduces Personal Intelligence to Connect Google Apps
