Google launches open AI translation models challenging ChatGPT Translate
16 Jan 2026

Google has unveiled TranslateGemma, a new set of open translation models based on Gemma 3.

The move is aimed at expanding multilingual support and providing developers with an open alternative to proprietary translation systems.

The TranslateGemma family comes in three model sizes—4B, 12B, and 27B parameters—covering everything from mobile devices to cloud deployments.

TranslateGemma's training process and performance
Model development

Google says the TranslateGemma models are trained using a two-stage approach. This includes supervised fine-tuning and reinforcement learning, leveraging high-quality human and synthetic translation data.

The company claims these models outperform baseline Gemma models by reducing translation error rates across high-, mid-, and low-resource languages while using fewer parameters.

TranslateGemma's open approach and local deployment
Accessibility

Unlike ChatGPT's translation feature, which is a closed system, TranslateGemma offers open weights. This means developers can download, inspect, and fine-tune the models as per their needs.

The open nature of TranslateGemma also allows it to run locally on devices, private servers or specific hardware setups without sending data to external servers.

TranslateGemma's support for multiple languages and multimodal capabilities
Language support

TranslateGemma supports 55 evaluated language pairs and has been trained on nearly 500 additional ones for future experimentation.

The models also retain the multimodal capabilities from Gemma 3, enabling text translation within images without separate training.

This feature allows developers to extend translation use cases beyond plain text, including image-based content and documents.

TranslateGemma's availability and Google's commitment to open translation
Model access

TranslateGemma models are available via Kaggle, Hugging Face, Google Colab, and Vertex AI.

Google has also released a technical report detailing the training methods, benchmarks, and supported languages for these models.

With this launch, the tech giant is positioning open translation models as a viable alternative to proprietary AI translation tools.