Which company recently launched TranslateGemma, a new open suite of AI translation models?

Google launched TranslateGemma on 15 January 2026 as an open-source AI translation model suite.

TranslateGemma is built on Gemma 3 architecture to enable high-quality multilingual translation.

The models support 55 global languages and nearly 500 additional language pairs.

TranslateGemma is available in 4B, 12B, and 27B parameter sizes for mobile, local, and cloud deployment.

The 12B model outperforms the larger 27B Gemma 3 baseline in translation efficiency.

Smaller models maintain state-of-the-art translation quality while reducing computational cost.

Performance is achieved through two-stage training: Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL).