Gemma 3n
Google DeepMind🇺🇸 USFrontier· Mar 1, 2026
Visionopen weights✎ Editorial
Google's on-device multimodal model: handles text, image, video AND audio at just 4B effective parameters. MatFormer architecture enables sub-models within the same checkpoint. Designed for edge deployments.
Editorial
- Context
- 33K tokens
- Params
- 8B raw (4B effective)
- License
- Gemma Terms of Use
- Modalities
- text, image, video, audio → text
1 variant
| E4B | 8B raw (4B effective) | 33K | HF |
Sources
HuggingFaceBlogDeutsch
Googles Multimodal-Modell für Endgeräte: verarbeitet Text, Bild, Video UND Audio bei nur 4B effektiven Parametern. Die MatFormer-Architektur ermöglicht Teilmodelle innerhalb desselben Checkpoints. Für den Einsatz am Edge ausgelegt.
Provenance
- Detected
- Mar 1, 2026
- First source
- kuratiert
- Description
- Editorial
- Editorially reviewed
- yes