Gemma 4 12B Locally: Google's Most Powerful Local Model Now in Commander Flow

Google has just released Gemma 4 12B — its newest and strongest model for running right on your computer. We have already added it to Commander Flow — as one more model to choose from, alongside the earlier ones.

And trying it is very simple. Download the app, install it, launch it — nothing complicated, it all works right away. And if you are already a customer, all you need is to restart the app: it updates itself, and the new model shows up in the list.

What is Gemma 4 12B

It is Google's fresh flagship for local AI — 12 billion parameters, and yet it was built to run not in someone else's data center but right on user devices. That is exactly why you can run Gemma 4 12B locally. What it can do:

• Multimodal out of the box — it understands not only text but also images, audio and video.
• Huge context memory — it keeps whole documents in mind, not just the last paragraph, and does not lose the thread in a long conversation.
• Fast — there is a built-in acceleration mechanism inside, so answers appear almost instantly, with no loss of quality.
• Strong at reasoning and code — a noticeable jump over the previous generation, with a "thinking" mode for complex tasks.
• Careful with languages — English, Russian and a dozen other languages with no machine accent.

And all of it — on your own hardware. Not a single line of your text goes to the cloud, gets passed to anyone, or is stored anywhere on someone else's servers.

Right under your cursor

There are two ways to work with Gemma 4 12B — and neither one takes you out of your current window.

A local chat on a separate hotkey. Press the second hotkey and a chat with the model opens. You talk to it exactly the way you would with familiar ChatGPT: you ask questions, ask it to write something, discuss an idea. Only all of it happens locally, on your computer, and nothing goes outside.

Operations on selected text — right under your cursor. Highlight any text in any app and ask to do something with it. For example, highlight a contract paragraph and ask to rewrite it tilted toward the client — or, the other way around, toward the contractor. The model returns a finished version on the spot, with no copying and no extra tabs.

And because the model lives on your computer, you can safely trust it with your work correspondence, draft contracts, personal notes and the ideas you are paid for: your data stays only with you.

"Google's freshest model. On your computer. And yours alone."

We keep an eye on what's new

Our team tracks every fresh model release and builds the best of it into updates — so you don't have to figure it out yourself. Something new and strong comes out — it shows up right in the app.

Honest about resources

Gemma 4 12B is our most powerful and heaviest model: it needs a fair amount of RAM. So, honestly: it won't run on every computer. If the resources fall short, the model simply does not activate — nothing breaks, the app keeps calmly working on the lighter models. But it is absolutely worth trying — maybe your machine can handle it.

A tip: before launching the heavy model, free up memory — close unused programs and extra browser tabs. Often that alone settles the matter and gives the model enough resources to start.

If you like it

The main saving is in how you work every day. Instead of endless tapping on the keyboard, you press one button: you dictate text by voice and edit it by voice too. Your hands get a rest, and the text comes together many times faster.

And that easily translates into money. At an average pay of about €8 an hour, the time saved on typing works out to roughly €2,000 a year. For the symbolic price of a subscription you get a tool that pays for itself many times over — and saves hours every week.

We'd be glad to see you among our premium users. Google's newest model, on your computer, under your cursor — download it and try it yourself.

FAQ

What is Gemma 4 12B and how is it different from the earlier models in the app?

It is Google's newest 12-billion-parameter model built to run locally. In Commander Flow it is added as one more model to choose from — the most powerful and the heaviest one. It understands not only text but also images, audio and video, holds a large context and answers fast thanks to built-in acceleration.

What do I need to do to try the new model?

Download the app, install it and launch it — nothing complicated. If you are already a customer, just restart the app: it updates itself, and Gemma 4 12B appears in the list of models.

Will Gemma 4 12B run on my computer?

Not on every machine — it is the heaviest model and needs a fair amount of RAM. If the resources fall short, the model simply does not activate, and the app keeps working on the lighter models. Before launching the heavy model, free up memory: close unused programs and extra browser tabs.

Does my data go anywhere when I use the model?

No. The model runs locally, on your computer. Neither your work correspondence, nor draft contracts, nor personal notes are sent to the cloud or stored on anyone else's servers.

How can I use the model in the app?

Two ways. On a separate hotkey — a local chat: you ask questions and ask it to write something, just like in regular ChatGPT, only local. And operations on selected text: you highlight a fragment in any app and ask to rewrite it — for example, a contract paragraph tilted toward the client or the contractor.