Google DeepMind Open-Sources TranslateGemma: Efficient Translation Models That Beat Larger Baselines at Half the Size
DeepMind released TranslateGemma, a family of open translation models in 4B, 12B, and 27B parameter sizes supporting 55 languages — with the 12B model reportedly outperforming the 27B baseline.
Google DeepMind released TranslateGemma on Thursday, a new family of open-source translation models built on the Gemma 3 architecture. As @GoogleDeepMind announced, the models come in 4B, 12B, and 27B parameter sizes, support 55 languages, and are "designed for efficiency without sacrificing quality."
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.