Google has begun expanding access to the Gemini 4 Argon model inside the Gemini app, after initial access in September was limited to the AI Ultra tier, which starts at $99.99. The new rollout includes Google AI Pro and AI Ultra subscribers, along with other development services and tools within the company’s ecosystem.
What does Argon offer inside Gemini?
Google describes the model as offering advanced reasoning and better performance in complex workflows. According to the company, Argon responds more directly, reduces hedging, repetition, and formulaic phrasing, and follows style instructions more consistently than previous Gemini models.
Google says the model shows clearer and more consistent progress while carrying out lengthy tasks. It has also enabled Gemini app users on macOS to switch between conversation and coding, with direct access to the Antigravity agentic coding interface for building applications, web experiences, and more.
A new toggle also appears in the main conversation window for switching from chat to tasks, which were previously called Spark. The model selector for AI Pro and AI Ultra subscribers includes the 3.5 Flash-Lite, 3.8 Flash, 3.1 Pro, and 4 Argon models, and Argon is also available to these subscribers in AI Mode.
Availability varies across Google tools
In the Antigravity desktop app and the Antigravity CLI, access to Argon requires an AI Ultra subscription. The same requirement applies to AI Studio and Android Studio (Canary). Inside the Gemini API, the model is available at an introductory price of $2 per million input tokens and $10 per million output tokens through November 30, 2026. After that, the price will rise to $4 per million input tokens and $20 per million output tokens.
Google’s examples of programming use
Google provided examples of using Argon in its internal version of Antigravity. The company says the model helped the Chrome team separate web-page processing from an old directory containing around 6 million lines within the 40-million-line Chromium codebase, completing a refactor across hundreds of files in weeks rather than months. The company connects this work to the future ability to move web pages directly between tabs, sidebars, and floating windows without reloading, while eliminating an entire class of browser failures.
In another example, Google said Argon achieved a 2.85-fold speedup for end-to-end inference on an A100 using vLLM, increasing throughput from approximately 37,900 to more than 108,000 tokens per second on Gemma 4 E2B, with exact token matching and no decline in quality. It also helped halve the CPU footprint of YouTube’s comments interface; optimizing emoji rendering reduced comment loading time on the server by 30 milliseconds and full client-side response time by 20 milliseconds.
What changes in practice?
This rollout moves Argon from a limited experiment for Ultra subscribers to a model available to a broader segment of Gemini users, but it does not eliminate the differences between plans. Advanced development features in Antigravity, AI Studio, and Android Studio remain tied to AI Ultra, while the API allows users to try the model for a variable fee after the introductory pricing ends. Google also announced that it will give 10,000 top scientists six months of free access to the highest usage level in Google Antigravity, and has begun accepting applications.