Google has released Gemma 4, a family of models focused on local, on-device AI inference for Android development. The lineup includes three models: Gemma E2B (8GB RAM, 2GB storage), Gemma E4B (12GB RAM, 4GB storage), and Gemma 26B MoE (24GB RAM, 17GB storage). The 26B MoE model targets desktop coding assistance in Android Studio, enabling agentic coding without sharing code with cloud providers — ideal for privacy-sensitive environments. The two smaller models target on-device inference, with E2B offering 3x faster inference than E4B. All models are up to 4x faster than previous versions and use up to 60% less battery. Gemma 4 also serves as the foundation for the next Gemini Nano generation. Developers can access the models via the AICore Developer Preview, Ollama, or LM Studio.
Sort: