Intel / post

Gemma spinoffs land for medical reasoning and 55-pair translation

Source: x.com
Date: Published Jan 16, 2026

This week, we released a pair of Gemma updates that developers can use to innovate across the healthcare and translation spaces: — MedGemma 1.5 4B: Our updated open medical AI model improves performance in 3D imaging and reasoning over medical records. We also kicked off the MedGemma Impact Challenge on @Kaggle to inspire development of AI applications that can help patients — TranslateGemma: This collection of multimodal open translation models is built on Gemma 3 and enables developers to build translation tools across 55 language pairs. The models are efficiently compact (available in 4B, 12B, and 27B parameter sizes), achieving high-fidelity translation quality with lower latency You can access both Gemma models on @huggingface, MedGemma on Vertex AI, and TranslateGemma on Kaggle. To learn more about Gemma and our latest breakthroughs from @GoogleDeepMind, visit https://t.co/Mf9nPsX2vA!

Jan 16, 2026 View on X

Context

Google AI's January Gemma updates give developers specialized starting points for medical imaging and translation. MedGemma 1.5 4B accepts multiple CT or MRI slices, allowing applications to work with three-dimensional scans. Its release guidance requires validation and adaptation for the intended medical use; the model's outputs are not intended to directly determine patient care. Medical-model announcement.

TranslateGemma specializes Gemma 3 using human and synthetic translations, followed by training with translation-quality rewards. The team reports that its 12B model outperforms the 27B base model on the specified WMT24++ . That illustrates the purpose of specialization: improve a defined task without simply choosing a larger general model. Translation-model announcement.

Terms in this piece · Glossary

eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.

Read the source x.com