Google's Gemma 4 Runs On-Device on iPhone 17 Pro at 40 Tokens Per Second — With Image Understanding
A viral demo shows Google's Gemma 4 E2B model running locally on an iPhone 17 Pro with MLX optimization, achieving ~40 tokens/second and full image understanding with no cloud connection required.
A demo that went viral Saturday shows Google's Gemma 4 E2B model running entirely on-device on an iPhone 17 Pro, hitting approximately 40 tokens per second with full image understanding capabilities. As @adrgrondin showed in a video walkthrough, the model runs through MLX optimized for Apple Silicon — meaning no cloud roundtrip, no Wi-Fi dependency, and no API costs.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.