Gemini 4 Argon: what Google released and who can use it

Google's newest model has a name that sounds like a lab code, and people are searching for it because the company is now showing what it can do. Gemini 4 Argon went to developers on 13 October 2026 [1]. The first public demonstrations came two weeks later, on 27 October, when Google posted eight examples of the model at work [2].

What developers got on 13 October

Google describes Gemini 4 Argon as its frontier model for complex workflows, with an emphasis on real-world software engineering. The company also points to enterprise knowledge work and cybersecurity defence [1]. Developers can use it on the paid tier of the Gemini API in Google AI Studio and Android Studio. Google AI Ultra subscribers can also try it in AI Studio and in Google Antigravity, the company's platform for agent-based development [1].

The pitch is about long tasks rather than quick answers. Google says the model is built for long-horizon workflows that need deep reasoning and structural precision [1]. That is a claim about design, not a benchmark result, and the company has not published independent test numbers in the pages we checked.

What the eight examples show

The 27 October post is mostly a set of demos, and each one is a single prompt or a single project. In one, a short prompt with no assets produced about 5,000 lines of maths code that rendered a 3D library, and the model then made a video about the process [2]. In another, Google fed it more than 80,000 raw NASA files together with a slide template, and it produced a deck an executive could present [2].

Some examples are more playful. Google used photos of a physical desk to build a playable 3D track, with the model coordinating subagents and a Blender connection in Antigravity [2]. A separate demo turned a PDF of academic chapters into an illustrated animation inside the Gemini app [2]. Google also showed a self-navigating 3D periodic table with atoms that move in time with audio [2].

The examples come from the company that built the model, so they show what Google chose to show. They do not measure how often the model gets things right, and the pages we checked contain no independent test.

Who gets it next

On 27 October Google said that AI Pro and Ultra subscribers can use Gemini 4 Argon in the Gemini app and in AI Mode in Search [2]. That widens the audience from developers to people who never open an API console.

What to watch next

Two things will tell readers more than the demos. The first is how the model does on ordinary tasks once people outside Google start using it. The second is pricing and limits for the Gemini app, which the pages we checked do not state. Until Google publishes those figures, anyone who pays for the model should check the current rates in AI Studio before building on it.