On-device AI has been a talking point for years, but Google’s latest move makes it harder to dismiss. Gemma 4, Google’s open-source model family, now runs directly on iPhones, full local inference, fully offline. It’s a meaningful step, and it signals that edge AI deployment isn’t a future priority anymore; it’s happening right now. So, where does Gemma 4 stand against the competition? Early benchmarks position the 31B variant alongside Qwen 3.5’s 27B model, a reasonably close matchup, with Gemm...