TestingCatalog claims clues to Gemini 4 Argon rollout and separate Carbon checkpoint testing
TestingCatalog AI News ● Covered by 8 sources
Google’s Gemini 4 Argon is showing up in Antigravity and Web. There are signs of a wider rollout, plus another test model called Carbon.
Based on reporting by TestingCatalog AI News — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google looks to be moving Gemini 4 Argon toward a broader release. Fresh Antigravity changes now reference the model in the selector, and they come with three context options: 256K by default, 512K at about 1.3x quota per turn, and 900K at around 1.8x quota per turn. Those settings look a lot like what Google already offers for Gemini Flash models, though the exact link between context size and reasoning effort is still murky.
The timing fits with Google’s own announcement from September 30, when it said Argon would first go to selected cybersecurity partners. Paid API customers and Google AI Ultra subscribers are next in line. Google hasn’t given a public date for a wider rollout, but the latest product changes suggest the machine is already being lined up for it.
There’s another shift happening in Gemini Web too. The service now supports low, medium, and high thinking-effort settings across available models. That kind of control would make sense if Google is preparing the platform for Gemini 4, even if the company hasn’t spelled out the whole plan.
Business Insider added another layer to the story: Google employees are apparently testing more Gemini 4 checkpoints internally. One of them, Carbon, is said to feel comparable to Claude Opus 5.5 for coding. The report says Barium-B is the checkpoint chosen for the public Argon release, while Carbon is a separate, potentially stronger version being tested through Jetski, Google’s internal name tied to Antigravity. All of that points to a company that isn’t just shipping one model. It’s building a pipeline of them.
My take — AI-written commentary, not fact-checked reporting
Google is doing the classic platform move: show one model, test three more, and keep everyone guessing about which one is actually the good one. The real story here is not the shiny rollout; it’s the quiet obsession with autonomous coding agents and tighter control over how much brainpower each turn gets. That’s very Google: promising simplicity, then hiding a small operating system under the hood.
Read more about this at: TestingCatalog AI News
