Gemini 4 Argon opens to developers and paying app users
Google has broadened access to Gemini 4 Argon after its restricted Fairwind rollout. Developers can use the API, while Pro and Ultra subscribers gain access in the Gemini app and AI Mode. Antigravity access remains tied to Ultra. The launch combines introductory API pricing with safeguards Google describes for monitoring misuse and agent behaviour, extending access beyond selected cybersecurity partners.
Artificial Intelligence··Night
Argon reaches APIs, the app and search
Google opened Gemini 4 Argon to developers and paying consumer subscribers on October 13, broadening a release previously restricted to selected partners. The Gemini API, an application programming interface for connecting software to the model, is available through AI Studio and the Gemini platform. Google AI Pro and Ultra subscribers can use Argon in the Gemini app and AI Mode, the conversational search service. Access therefore spans software developers and people using Google’s consumer interfaces.[1], [2], [3]
The access conditions depend on the product being used. Google specifies Ultra membership for trying Argon in Antigravity, its agent-based software development environment. The 9to5Google account likewise names Ultra for the desktop application and command-line tool, while describing Pro and Ultra access in the Gemini app. It also reports Ultra requirements in AI Studio and Android Studio Canary. Those product-specific subscription statements sit alongside Google’s announcement of general paid API access.[1], [2]
Introductory API prices end in November
The introductory API schedule charges 2 dollars per million input tokens and 10 dollars per million output tokens, according to both 9to5Google and Engadget. Tokens are the small units of text the model processes and produces. The launch prices apply through November 30; the announced subsequent charges are 4 dollars for input and 20 dollars for output on the same million-token basis. This schedule prices incoming material separately from the model’s generated answer. Google states that cached input tokens receive a 95 per cent discount.[2], [3], [1]
Google identifies long-running coding, legal, finance and cybersecurity work among the intended uses. It directs developers to a continuation protocol for extended tasks and states a combined input-output limit of one million tokens. Company examples include reorganising legacy Chrome code and improving Python session preparation in its Agent Development Kit. These examples describe work by Google’s teams; the broader rollout makes the model accessible through additional products and accounts.[1]
The wider release retains a separate cybersecurity programme
Google’s September 30 announcement described an initial rollout to trusted cyber defenders through Fairwind, its restricted partner programme. The October 13 announcement expands public access while also beginning an extension of Fairwind to hundreds of additional partners. Applications remain necessary for the separate version without cyber guardrails. The chronology moves from selected defenders to developers and paying app users, with the partner route continuing alongside the wider launch.[1], [3]
For the broader model, Google describes misuse monitoring, enforcement and resistance to prompt injection, where outside instructions try to divert an agent’s task. It says a monitor examines agent reasoning and actions in internal use and evaluations. These are the company’s stated safeguards. Google also opened applications for eligible scientists to receive six months of access to Antigravity’s highest usage tier, subject to capacity and selection.[1]
