- Google introduced TranslateGemma, translation models supporting 55 languages with 4B, 12B, and 27B parameters
- TranslateGemma 12B outperformed the larger 27B Gemma 3 on the WMT24++ benchmark with fewer parameters
- Models provide efficient, high-quality translation for smartphones, laptops, and cloud systems
Google has unveiled TranslateGemma, an open suite of translation models based on its Gemma 3 architecture, offering text translation in 55 languages. The models, which come in 4B, 12B and 27B parameter versions, are designed to balance translation accuracy with efficiency, allowing them to run on everything from smartphones to consumer laptops and cloud systems.
Google said evaluations indicated that TranslateGemma's 12B model outperformed the larger Gemma 3 27B benchmark model on WMT24++, while requiring significantly fewer parameters to do so. WMT24++ is an expanded benchmark dataset for evaluating the performance of machine translation systems and large language models (LLMs) across a wide variety of languages.
The assessment also pointed to stronger results across languages with different levels of data availability and improved handling of text translation within images.
“For developers, this is a massive win. You can achieve high-fidelity translation quality using less than half the parameters of the baseline model. This efficiency breakthrough allows for higher throughput and lower latency without sacrificing accuracy. Similarly, the 4B model rivals the performance of the larger 12B baseline, making it a powerful model for mobile inference,” Google said in a blog post.
The launch is intended to give researchers and developers access to open, flexible tools for creating efficient translation systems and strengthening multilingual communication.
“We have rigorously trained and evaluated TranslateGemma on 55 language pairs, ensuring reliable, high-quality performance across major languages (such as Spanish, French, Chinese, and Hindi) as well as many low-resource languages. Beyond these core languages, we pushed the boundaries by training on nearly 500 additional language pairs,” Google said.
According to Google, developers can obtain the TranslateGemma models through multiple outlets, including Kaggle, Hugging Face, Vertex AI and the Gemma Cookbook.
“TranslateGemma models retain the strong multimodal capabilities of Gemma 3. Our tests on the Vistra image translation benchmark show that the improvements in text translation also positively impact the ability to translate text within images, even without specific multimodal fine-tuning during the TranslateGemma training process,” Google added.
In parallel, OpenAI has launched ChatGPT Translate, a web-based translation product designed to compete with Google Translate. The service uses ChatGPT's underlying models to handle translations across over 50 languages and includes automatic language recognition, while optional sign-in unlocks additional functionality.
The interface also includes a set of one-tap options at the bottom of the screen, allowing users to quickly refine a translation. These shortcuts offer choices such as improving fluency, shifting the tone to business formal, simplifying the language for younger readers, or adapting the text for academic use. Selecting an option transfers the text to the main ChatGPT interface, where users can further customise it using generative AI.
Watch LIVE TV, Get Stock Market Updates, Top Business, IPO and Latest News on NDTV Profit.