A real AI model now fits in 1 GB and runs on your laptop

Here's a neat one. Usually an AI model lives on someone else's server, and you reach it over the internet. But on June 5 Google released versions of Gemma 4 that fit right on your laptop.

What happened

Google took its open Gemma 4 model and applied a trick called QAT — the model is trained to "think" in a compressed form from the start, so it barely loses quality when shrunk.

The numbers: the smallest version (Gemma 4 E2B) shrank to under 1 GB. That's the size of a couple of phone photos — with a working language model inside.

And running it isn't hard. Google shipped ready builds for popular free apps: Ollama, LM Studio, llama.cpp and others. Install the app, download the model — and it answers right on your machine, no server.

What's in it for you

This is the cheapest way to touch AI "for real": the model runs locally, for free, with no internet. No tokens, no bills — your laptop does the work itself.

Want to try? Install LM Studio or Ollama, pick Gemma 4 E2B, and type it something. You'll also see that a model is just a file you can download.

No hype: your own AI model in your pocket stopped being sci-fi — it's now under a gigabyte.

Source: Google, Android Authority