Google has begun rolling out Gemini 4 Argon in the Gemini app and in Antigravity, but only for paying users. The model, presented in late September as a tool for coding, enterprise work and cyber defense, had until now been reserved for a small group of partners, largely because of concerns about malicious use. Access is now broader, but with clear limits.
In the Gemini app Argon arrives today for general queries and requires either an AI Pro subscription, at 20 dollars a month, or AI Ultra, starting at 99 dollars a month. This fits Google's recent move to restrict free accounts, which since October have lost Flash and Pro and keep only Flash-Lite. Pro subscribers can generate code with Argon in the app's Canvas mode, but the multi-threaded agentic workflows in Antigravity and AI Studio are reserved for the Ultra plan, because they consume the most compute. Google has not said when, or whether, Argon will reach Antigravity on other subscription tiers. For the most demanding programming tasks the company is keeping Gemini 3.1 Pro available.
Developers using the paid API pay per token. The introductory price is 2 dollars per million input tokens and 10 dollars per million output tokens, with a 95% discount on cached tokens. That rate lasts until November 30, after which it rises to 4 dollars per million input and 20 dollars per million output. Anyone building workflows on Argon will need to factor that in.
Image: Google
The model at work inside Google
Google says thousands of its employees already use Argon. Among the examples it cites, the model reportedly made YouTube's comment system more efficient: removing redundant emoji rendering cut server load by 45%. On the proprietary drone physics simulator used in Project Wing, it reportedly found critical bugs that had been in the codebase since 2018. Robotics also benefits, with a measurable step up in embodied reasoning. In one demonstration, a humanoid robot named Apollo learned to play cornhole from a single human video, building its own physics simulator and landing seven throws in a row after training on the real robot.
Answers in the app also change in tone: Google says Argon responds more directly and with less hedging, repeats itself only when needed, and shows progress updates more often on longer jobs. The model can work for longer thanks to a unified limit of 1 million input-output tokens in the API.
Back in the frontier race
Argon is Google's first major release since Gemini 3 in late 2025, arriving at a time when OpenAI and Anthropic have pulled ahead. The launch follows a series of delays and internal reshuffles, including the postponement of Gemini 3.5 Pro, which slipped precisely because of coding shortfalls. In the same weeks Google is already testing internal variants that follow. Argon's release therefore does not close the race: it signals that the company wants to regain ground with a paid product, calibrated to compute costs before reaching a wider audience.
No comments yet. Be the first!