Gemma 4 Apache 2.0 License: What Open Source Really Means for AI Builders

News

Google Ships Gemma 4 Under Apache 2.0: The Licensing Story Matters More Than the Benchmarks

Apache 2.0 is a bigger deal than most coverage acknowledges. It means no custom terms to negotiate, no usage restrictions to lawyer-proof, and no surprises when you ship a product built on top of it. For enterprise builders who got burned by restrictive model licenses before, this is the practical unlock.

Gemma Nano 4 Is 4x Faster and Uses 60% Less Battery Than Its Predecessor on Android

Gemma 4 is the base model for the next Gemini Nano generation, rolling out to devices that already have 140 million installs. The performance improvements are substantial enough that on-device AI finally stops feeling like a compromise.

Analysis

Gemma 4 Model Family: E2B Through 31B Dense, Hardware Requirements, and Real Benchmark Positions

The 31B Dense is sitting at #3 on Arena AI's open model leaderboard and the 26B MoE activates only 3.8B parameters at inference, which dramatically changes the hardware math. If you have been waiting for a locally runnable model with frontier-adjacent performance, the specs are finally credible.

Tools

Gemma 4 in Android Studio Agent Mode: Local Agentic Coding Without API Keys or Quotas

Running complex agentic workflows locally via Ollama or LM Studio means zero API costs and no data leaving your machine. The hardware bar is real though, the recommended 26B MoE model needs 24GB total RAM, so check your setup before getting excited.

LiteRT-LM Runs Gemma 4 E2B Under 1.5GB RAM on Raspberry Pi and Qualcomm Edge Chips

Processing 4,000 input tokens across two agentic skills in under 3 seconds on a Qualcomm Dragonwing IQ8 is a real number worth tracking. IoT and robotics builders now have a credible path to on-device agents without cloud dependencies.