Can I run Gemma 3 1B?

Gemma 3 1B is a tiny chat model from Google, built for the cases where you want a local assistant that loads instantly and runs almost anywhere. At a 4-bit quant it is only about 0.6 GB, and even an 8-bit build is roughly 1.1 GB, so it fits in 3 GB of RAM with room to spare. That means it runs on an old laptop, a Raspberry Pi class device, or a phone-grade chip without touching a discrete GPU. If you need something for quick chat, drafting, or embedding in an app where every megabyte counts, this is the model to reach for.

In daily use the headline is raw speed. On an RTX 4090 it pushes around 1413 tokens per second, and a more typical RTX 3060 still does about 505 tok/s, with an Apple M Max around 575 tok/s. Even pure CPU on DDR5 manages roughly 84 tok/s, faster than you can read. The context window is 32K, and unlike bigger models you can fill a good chunk of it cheaply: at the full 32K the whole thing sits around 2.2 GB total, so memory is never the thing that stops you here.

Honesty about its size matters: at 1B parameters this is not a reasoning or coding model, and it will lose the thread on long multi-step instructions. If you have the headroom, Gemma 3 4B generally handles harder prompts and adds vision, while Llama 3.2 1B is the obvious same-weight alternative to compare against for plain chat. Gemma 3 1B's standout trait is that it is about the smallest genuinely usable chat model you can run. One caveat: it ships under Google's Gemma license, which is open-weight but carries provider-specific terms, so read them before shipping it commercially.