Google: Gemma 3n 4B
Model Type
Open Weight Model
4B parameters
Recommended Use Cases
Text Generation
Try Gemma 3n 4B
Gemma 3n is Google's mobile-first open model optimized for on-device AI, featuring multimodal capabilities (text, image, audio, video) while running on as little as 2GB of RAM.
Meet Gemma 3n, a model that runs on as little as 2GB of RAM. It shares the same architecture as Gemini Nano.
- Google AI
Overview
Gemma 3n is a next-generation open model designed specifically for on-device deployment on phones, tablets, and laptops. Created in collaboration with Qualcomm, MediaTek, and Samsung, it shares architecture with Gemini Nano and introduces groundbreaking efficiency innovations that enable frontier-level AI directly on consumer devices.
Model Variants
| Model | Effective Params | Raw Params | Memory | Modalities |
|---|---|---|---|---|
| Gemma 3n E4B | 4B effective | 8B raw | ~3GB | Text, Image, Audio, Video |
| Gemma 3n E2B | 2B effective | 5B raw | ~2GB | Text, Image, Audio, Video |
Key Features
- Mobile-first architecture: Engineered from the ground up for edge devices
- True multimodal: Native support for text, image, audio, and video inputs
- Per-Layer Embeddings (PLE): Caches parameters to storage, dramatically reducing RAM usage
- 140+ language support: Text understanding in 140 languages, multimodal in 35 languages
- Offline operation: Full functionality without cloud connectivity
- Privacy-preserving: All processing happens locally on-device
Use Cases
- Real-time voice assistants and transcription
- On-device translation and language understanding
- Visual scene understanding and image analysis
- Privacy-sensitive applications
- Offline-capable intelligent features
Role in Series
The Gemma family offers models optimized for different deployment scenarios:
- Gemma 3 270M: Ultra-compact for task-specific fine-tuning
- Gemma 3n E2B/E4B: Mobile-first with 2-3GB RAM, full multimodal including audio (this model)
- Gemma 3 1B-27B: Cloud/desktop deployment with maximum capability
- CodeGemma: Code-specialized variant for development tasks