Google AI Edge Gallery debuts on macOS
▼ Summary
– Google released the Google AI Edge Gallery app for macOS, allowing local AI model use, alongside the Gemma 4 12B model and the Google AI Edge Eloquent dictation app.
– Local models like Gemma run on a device’s own processing power, do not need an internet connection, and offer greater privacy than cloud-based models.
– Google AI Edge Gallery for Mac currently only provides access to five of Google’s own models, unlike platforms like Ollama and LM Studio that support many open models.
– The new Gemma 4 12B model is multimodal and can run locally on laptops with 16GB of RAM, delivering performance comparable to larger models.
– Google AI Edge Eloquent is a free Mac dictation app that transcribes speech, polishes text, and processes everything on-device, with options for custom words and writing styles.
Google is expanding its local AI ecosystem with a trio of new releases for Mac users: the Google AI Edge Gallery app now available on macOS, the launch of the Gemma 4 12B model, and a new dictation tool called Google AI Edge Eloquent. Each brings on-device intelligence to Apple computers, emphasizing privacy and offline functionality.
To understand why this matters, consider how most people interact with large language models today. The dominant tools, like ChatGPT, Claude, and Gemini, rely on cloud servers. They are powerful but require an internet connection and send data to external data centers. Local models, by contrast, are smaller and less capable than their trillion-parameter cloud counterparts. Yet they offer distinct benefits: they work offline, run entirely on your computer’s hardware, and keep all conversation data private. The faster your machine, the quicker the response and the larger the model it can handle.
Until now, running local models on a Mac meant using third-party platforms like Ollama or LM Studio, then downloading a compatible model from a repository like Hugging Face. Google’s new Google AI Edge Gallery for macOS offers a more curated path, though with a key limitation. Unlike the open platforms, Google’s app currently supports only five of its own models. These include Gemma-4-12B-it, Gemma-4-E2B-it, Gemma-4-E4B-it, Gemma-3n-E2B-it, and Gemma-3n-E4B-it, where “it” denotes instruction-tuned versions that follow user prompts rather than simply complete text.
The standout is the Gemma 4 12B model, also released today. Google describes it as designed to bring “agentic, multimodal intelligence directly to your laptop.” While most consumer local models from frontier labs range from 2 billion to 9 billion parameters, Gemma 4 packs 12 billion. Google claims this size delivers performance comparable to its 26-billion-parameter mixture-of-experts model, yet remains small enough to run on consumer laptops with just 16GB of RAM. It is also multimodal, handling text, vision, and audio, and Google says it offers strong coding capabilities, enabling users to extract insights from data right on their device.
Alongside the Gallery and the new model, Google introduced Google AI Edge Eloquent for Mac. This free dictation app, which debuted on iOS months ago, captures speech and transcribes it while polishing the text. It removes disfluencies, makes light edits for clarity, and processes everything on-device rather than in the cloud. Users can choose between different writing styles and add custom words, such as names and jargon, to reduce the frequent miscorrections typical of dictation software.
For those interested, more details on Google AI Edge Gallery and Gemma 4 12B are available on Google’s official pages, along with information on the Eloquent app.
(Source: 9to5Mac)
