REDDIT · REDDIT// 9d agoMODEL RELEASE

Gemma 4 edge model beats Qwen 3.5

Google’s Gemma-4-E2B-IT delivers 4B-class performance in a 2.3B parameter on-device package. The Apache 2.0 multimodal model features native audio support and a fast-reasoning "thinking mode" optimized for mobile and edge hardware.

// ANALYSIS

Gemma 4 is a masterclass in parameter efficiency, proving architecture beats raw scale for mobile deployment.

–Native multimodal support (audio, image, video) outclasses text-only 4B competitors like Qwen 3.5
–Apache 2.0 license marks a major shift toward true open-source for Google’s small model family
–"Thinking mode" provides structured reasoning without the massive token bloat of traditional reasoning models
–Per-Layer Embeddings (PLE) and hybrid attention enable a 128K context window on consumer hardware

// TAGS

gemma-4gemma-4-e2b-itllmmultimodalreasoningopen-sourceedge-ai

DISCOVERED

9d ago

2026-04-03

PUBLISHED

9d ago

2026-04-03

RELEVANCE

10/ 10

AUTHOR

ZootAllures9111