Google brought its AI Edge Gallery to macOS on Wednesday, giving Mac users a first-party option for running Gemma models locally. The app joins a field already crowded with tools like Ollama and LM Studio, but with one key difference: Google's gallery only runs Google's models. The macOS launch follows versions for Android and iOS that have been available for months. Users download the app directly from Google's website and can run five instruction-tuned models without an internet connection. The list includes Gemma-4-12B-it, Gemma-4-E2B-it, Gemma-4-E4B-it, Gemma-3n-E2B-it, and Gemma-3n-E4B-it.
Competing platforms like Ollama and LM Studio let users install any model compatible with their hardware from thousands available on Hugging Face. Google's app is a curated experience. You get Google's models or nothing.
The flagship model in the lineup is Gemma 4 12B, a 12-billion-parameter open model Google says delivers performance comparable to its 26-billion-parameter mixture-of-experts model. It runs on consumer laptops with 16GB of RAM, which covers all modern Apple silicon Macs (the MacBook Neo is the notable exception, according to AppleInsider).
Gemma 4 12B is multimodal, handling text, vision, and audio. Google also touts strong coding abilities, letting users extract insights from local data without sending anything to external servers.
Alongside the gallery, Google released AI Edge Eloquent, an on-device dictation app that transcribes speech, removes filler words, and polishes text. It works across all Mac apps, launches via a keyboard shortcut, and processes everything locally.
Users can pick writing styles and add custom vocabulary for names and jargon. The app is free but currently English-only, with more languages promised. The timing matters. Local AI is gaining traction as users weigh privacy and offline access against the convenience of cloud models like ChatGPT, Claude, and Gemini.
Running models on-device means no data leaves the computer, no internet connection is required, and response speed depends on local hardware rather than server latency.
Google's move onto the Mac with a curated, first-party tool suggests the company wants a stake in that shift. But by limiting Edge Gallery to its own models, Google is betting users will trade flexibility for a smooth, Google-controlled experience.
Whether that trade-off resonates depends on how well Gemma 4 12B performs against the open ecosystem alternatives already on users' Macs.
