Google Drops Gemma 4 Under Apache 2.0 — Now Runs on a Raspberry Pi

Google releases Gemma 4 — a family of AI models under Apache 2.0. 31B model, MoE architecture, edge models on 1.5GB RAM.

Google··2 min read

The Big Move

Let's be honest: when Google releases its models under Apache 2.0, it's not charity. It's a dominance play. Previous Gemma releases came with restrictive licenses. Apache 2.0 erases all barriers.

The family spans four models:

  • Gemma 4 31B — dense model, ranked #3 on open leaderboards
  • Gemma 4 26B-A4B — Google's first mixture-of-experts: 128 experts, only 3.8B active parameters
  • Gemma 4 E4B and E2B — phones, Raspberry Pi, Jetson Nano. E2B runs under 1.5GB RAM

Technical Specs

Context windows hit 256K for large models, 128K for edge. Native function-calling, structured JSON output, and system instructions are built-in. The models support 140+ languages natively.

NVIDIA is already optimizing Gemma 4 for local RTX AI. Ollama added all four variants. You can run the 26B MoE locally in one command via llama.cpp.

The Numbers

Gemma has now passed 400 million downloads and 100,000+ community variants. The E2B and E4B are the foundation for Gemini Nano 4, shipping to Android devices later this year.