REDDIT · REDDIT// 5d agoMODEL RELEASE

Google DeepMind launches Gemma 4 under Apache 2.0

Google DeepMind has released Gemma 4, a new family of open multimodal models featuring a flagship 31B dense model and a hyper-efficient 26B MoE variant. The launch marks a major shift for the series by adopting a fully permissive Apache 2.0 license and introducing native multimodal capabilities with advanced reasoning for agentic workflows.

// ANALYSIS

Gemma 4 going fully Apache 2.0 is a massive strategic play by Google to dominate the open-source ecosystem against Meta's Llama series.

–The 26B MoE model activates only 3.8 billion parameters during inference, bringing frontier-class intelligence and extreme low latency to standard consumer GPUs.
–Native multimodal support across the family—including audio input on the smaller edge models—positions Gemma 4 as a complete, out-of-the-box foundation for complex autonomous agents.
–A staggering leap in reasoning and coding capabilities (jumping to Codeforces 2150) proves Google's distillation pipeline from the main Gemini models is yielding massive dividends for the open community.

// TAGS

gemma-4open-sourcellmmultimodalagentreasoning

DISCOVERED

5d ago

2026-04-06

PUBLISHED

5d ago

2026-04-06

RELEVANCE

10/ 10

AUTHOR

jacek2023