Product Launchtranslation modelsmultilingual translationgoogleon device inference
Google Releases TranslateGemma Open Translation Models
9.2
Google released TranslateGemma, an open suite of translation models based on Gemma 3, offering 4B, 12B, and 27B-parameter variants supporting translation across 55 languages. Trained with supervised fine-tuning plus reinforcement learning using automatic reward signals, the models achieved improved parameter efficiency on the WMT24++ benchmark—reportedly the 12B model outperformed a 27B Gemma 3 baseline—and target mobile to cloud deployments.
Key Points
- 1Introduce TranslateGemma: open 4B, 12B, and 27B Gemma 3 models.
- 2Optimize translation quality with supervised fine-tuning plus RL using MetricX-QE and AutoMQM rewards.
- 3Enable low-latency, cost-sensitive on-device and consumer-laptop translation with near-bigger-model performance.
Scoring Rationale
High novelty and usability from Google's official open release; strength in efficiency, limitation in scope to translation-specific models.
Sources
Public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
Active Search Campaigns by BudgetEasyHigh CPC Clicks & Poor Landing PagesMediumCampaign ROAS by Attribution ModelHard
250 free problems · No credit card
See all Ad Tech problems
