Today's signal
Google launched AI Edge Gallery — a free, open-source app that runs Gemma 4 large language models directly on your phone. No internet. No cloud. No subscription. It hit #8 on the App Store's productivity charts within days. Available on both Android and iOS right now.
Why it matters
Until now, using a powerful LLM meant paying for a subscription or burning API credits through someone else's servers. Google just removed that entire equation. The Gemma 4 models running inside this app support 128K context windows, multimodal input (text, images, audio), and over 140 languages — all processed on-device with under 1.5GB of memory. Your prompts never leave your phone. It even has Agent Skills, one of the first implementations of multi-step agentic AI running fully offline on a consumer device. Google built this with Qualcomm and MediaTek, optimized for the latest mobile AI chips. The app is open-source under Apache 2.0, meaning anyone — including businesses — can use, modify, and deploy these models without restriction.
👉 Download Google AI Edge Gallery: App Store | Google Play
The take
This is Google's real AI play — not Gemini, not the cloud upsell. This is the move that makes every other AI subscription look like a tax on people who didn't know better. When a frontier-class model runs for free on the phone already in your pocket, the question stops being "which AI do I pay for?" and starts being "why am I paying at all?" OpenAI charges $20/month. Google just made the counter-argument downloadable.
The number
#3 — where Google's Gemma 4 31B Dense model ranks among all open models on the Arena AI chat leaderboard, outperforming models 20x its size.