Google opens Gemini 4 Argon to AI Pro subscribers and keeps Antigravity's own apps on Ultra

A developer on a Pro plan opens the Gemini app on a Mac and finds four models in the picker: 3.5 Flash-Lite, 3.8 Flash, 3.1 Pro and 4 Argon ^[3]. The same developer then opens the Antigravity desktop app to try the workflow Google has been showing off, and the new model asks for Ultra ^[4].

The easy summary is that Pro now has Argon for coding. The access list supports less. Pro and Ultra both get Argon in the Gemini app ^[1], and the Mac app adds Antigravity's coding interface, which Google describes as letting users "switch between chatting and coding" ^[17]. 9to5Google's report does not say whether that interface runs Argon for Pro subscribers. It does say that Argon in the Antigravity desktop app and CLI requires Ultra ^[4].

Google's own examples sit on the Ultra side of that line. The company said it produced them with its internal version of Antigravity ^[13]. In the Chrome case, Google says Argon helped separate core webpage handling from a roughly 6 million line legacy folder inside Chromium's 40 million-line codebase, "completing months of work in weeks" ^[15]. For Gemma, Google reports a 2.85x speedup on an A100 via vLLM, from about 37,900 to over 108,000 tokens per second ^[14]. Google's benchmark claims against GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 are also its own ^[8]. Ars Technica wrote that nothing about the model's capabilities has changed with this rollout ^[9].

The API is metered, so it follows different rules from seats. Argon costs $2 per million input tokens and $10 per million output tokens until November 30, 2026, then $4 and $20 ^[6]. That is a doubling on both ^[18]. The Verge put the introductory price at around a fifth of what OpenAI's GPT-6 Astra and Anthropic's Fable 5 models cost, and said Argon stays significantly cheaper after the increase. It added that models use different numbers of tokens for the same task, so a per-token price does not settle the cost ^[16]. A team that measures cost per task before November 30 should double the result before it goes into a budget ^[18].

We'd start every candidate task on Pro and move it up only when the task forces it. Sort each task on two axes: whether it fits in a chat with the files attached or needs the whole repo, and whether it is a one-off or will recur. A chat-sized one-off is a Pro task today, in the Gemini app. A chat-sized task that will recur belongs in an API cost test at the post-November rate. For a repo-sized one-off, buy one Ultra seat for one person with an end date, since Ultra is what the Antigravity desktop app, CLI, AI Studio and Android Studio (Canary) require [c6, c7]. Repo-sized recurring work comes last, after a one-off has produced a result the team would accept in review. The tradeoff is that a Pro trial leaves the Antigravity desktop app and CLI untested, and Antigravity is where Google's examples were produced ^[13].