Fireship: Google Gemma 4 Under Apache 2.0 Disrupts Open-Source AI — 400M+ Downloads, Runs Offline on Edge Devices

Fireship·medium signal

Fireship analyzes Google's Gemma 4 release (1.2M views) — built from the same research as Gemini 3, released under fully permissive Apache 2.0. Models activate 2B/4B parameter MoE footprints during inference for RAM efficiency, with 256K context, native vision/audio, and 140+ language support. Run fully offline on phones, Raspberry Pi, and Jetson Orin Nano. Over 400M total Gemma downloads and 100K+ community variants make this the most significant open-weight release since Llama.