Gemma 3n E4B

Overview

Google's full Gemma 3n with 4B effective parameters (8B raw) and nested MatFormer architecture. Native multimodal across 140 languages for high-end mobile deployments.

When to pick this model

• High-end mobile or edge devices needing multimodal input
• Multilingual on-device assistants across 140 languages
• Image-aware mobile workflows
• Replacing E2B when accuracy matters more than RAM

VRAM requirements by quantization

Quantization	VRAM required
Q4_K_M (recommended)	4.5 GB
Q5_K_M	5.5 GB
Q8_0	8 GB
FP16 (no quantization)	14 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

Strengths

• 4B effective parameters punch well above mobile-class weights
• Integrated multimodal — text and image input
• 140 language coverage
• Open Gemma license

Limitations

• 32k context only
• Beaten by Gemma 3 12B in desktop scenarios
• Gemma license — less permissive than Apache 2.0
• Multimodal support uneven across runtimes

Architecture & training

Architecture: Gemma 3n E4B · on-device architecture · 4B effective

Training: Google Gemma 3n 4B, multimodal text+image, 140 languages.

Quick start

ollama run gemma3n:e4b

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.