Google rolled out its AI Edge Gallery for macOS, letting Mac owners run Gemini‑derived Gemma models locally. The launch bundles the new Gemma‑4 12B model and a free dictation tool called Google AI Edge Eloquent. Unlike cloud‑based ChatGPT or Claude, these models execute on the laptop’s CPU or GPU, preserving privacy and working offline.

Gemma‑4 12B packs 12 billion parameters, a size Google claims matches the performance of its 26 billion‑parameter mixture‑of‑experts system while fitting on laptops with 16 GB RAM. The multimodal engine handles text, images and audio, and includes coding assistance that can analyze data locally. Only five Google‑curated models are currently available through the macOS Gallery.

The companion Eloquent app transcribes spoken words on‑device, cleans up disfluencies and applies style presets, while letting users add custom vocabulary to avoid misrecognition. By keeping inference on the Mac, Google targets power users who demand speed, data control and offline capability, challenging the dominance of cloud‑only offerings. The rollout marks a clear shift toward local AI on consumer hardware.

Earlier this year Apple and other vendors saw third‑party tools like Ollama and LM Studio enable arbitrary model installs on Macs, but those require manual setup and internet fetches. Google’s Gallery simplifies the process by curating a small, vetted suite, though it restricts users to its own lineup. The approach may appeal to enterprises prioritizing secure, on‑prem AI workloads.