From smartphones to Raspberry Pi: Running Gemma 4 anywhere

Does every AI task need a massive cloud model? Join Muhammad Farooq (GDE) and Omar Sanseviero (Lead AI Developer Experience at Google DeepMind) live from Google Cloud Next '26 to discuss the record-breaking launch of Gemma 4. With over 40 million downloads in just three weeks, Gemma 4 has become the definitive open model family for developers who want intelligence without the massive footprint. In this deep dive, Omar explains why the shift to Apache 2.0 licensing and "Hybrid Inference" is changing the game for startups and regulated industries alike. Key Highlights: The Gemma Family: From the 2B parameter model (optimized for mobile) to the 31B parameter model (designed for consumer GPUs), learn how these models provide the highest "intelligence per parameter" on the market. Multimodal & Multilingual: Gemma 4 supports over 140 languages and offers on-device vision, video, and audio capabilities (2B/4B models), making it a truly global tool. The "Gemma-verse": How communities are already fine-tuning the model for specialized tasks, such as Quechua-to-Spanish translation for indigenous communities in Peru. Hybrid Inference & Local Routers: Discover the future of "Cactus Compute"—using local models like Gemma for 80% of daily tasks and only calling the cloud (Gemini) for high-complexity queries. Sovereign & Offline AI: Why Gemma 4 is the top choice for healthcare, government, and offline scenarios where data privacy and lack of internet are critical barriers. On-Device Agents: See how small models are now powerful enough to act as local agents—controlling device hardware like flashlights or drafting emails directly on Android. "Most people don't have a cluster of H100s. We designed Gemma so it can run on a Raspberry Pi, a Jetson Nano, or the phone in your pocket, without sacrificing the reasoning skills you need for agentic workflows." Download Gemma 4 on the Android AI Edge Gallery, experiment with the weights on Hugging Face, and join the Gemma-verse to s