Google Shrinks Its AI Model So Your Laptop Can Handle It

Google just released a new version of its Gemma 4 AI model that’s built for the average laptop. The company calls it Gemma 4 12B, and the big selling point is that it only needs 16GB of RAM to run locally. That’s a fraction of what most large language models require, and it means you don’t need a specialized, expensive machine to use it.

The original Gemma 4 family launched in April with four models: two small ones for mobile devices and two heavy-duty options—a 26-billion-parameter mixture-of-experts model and a 31-billion-parameter dense model. That left a big gap in the middle. The new 12-billion-parameter model fills that space. It’s far more capable than the mobile versions, but it won’t demand a $20,000 AI accelerator.

Google says the 12B model is almost as good as the 26B model on benchmark tests, while using roughly half the memory. That’s a meaningful trade-off for anyone who wants to experiment with AI on a regular laptop without upgrading their hardware. The company also notes that the model works with either system RAM or video RAM, so it’s flexible.

With this release, Google is acknowledging what many users have felt: the generative AI boom has pushed memory costs up, and not everyone can afford the high-end gear. Gemma 4 12B won’t replace the big models for heavy lifting, but for everyday tasks on a standard laptop, it might be enough.