Following a late-September announcement, Gemini 4 Argon is now rolling out in the Gemini app, Google Antigravity, and other platforms. In the Gemini app, Gemini 4 Argon is now available to Google AI Pro and AI Ultra subscribers. Last month, initial availability was limited to the Ultra ($99.99+) tier. Google says users can expect “frontier reasoning and improved performance on complex workflows” within the app.
In addition, “Argon answers questions more directly, reducing unnecessary ambiguity and avoiding repetition and clichés in its responses.” It also “follows style instructions more reliably than any previous Gemini model.” For long-running tasks, it “shares progress more frequently and effectively, keeping users informed at all times.” Given a “significant improvement” in coding performance, Google is allowing users to “switch between chat and coding” in the Gemini app on macOS. Users will be able to directly access Antigravity’s agentic coding interface to build apps, web experiences, and more.
Major user-experience improvements with Gemini 4 Argon
At the same time, the main chat window will gain a toggle button that lets users switch from chat mode to task mode (formerly called Spark). As of today, AI Pro and AI Ultra subscribers will see 3.5 Flash-Lite, 3.8 Flash, 3.1 Pro, and 4 Argon in the model selector. AI Pro and AI Ultra subscribers can also access Gemini 4 Argon in AI Mode.
In the Antigravity desktop app and Antigravity CLI, the new model requires AI Ultra. Today Google shared more examples of how Argon is being used in internal versions of Antigravity for Chrome, Gemma, YouTube, and more: Argon helped the Chrome team decouple core web processing from roughly 6 million lines of legacy browser feature code residing in Chromium’s 40-million-line codebase.
By handling multiple discrete code changes simultaneously, Argon helped the team complete architectural refactoring across hundreds of files in weeks — work that would typically take months. For Chrome’s billions of users worldwide, this will unlock the ability for live web pages to move fluidly between tabs, side panels, and floating windows without reloading, eliminating an entire class of browser crashes.
By using Argon, we achieved a 2.85× end-to-end inference speedup on A100 via vLLM (raising throughput from roughly 37,900 to over 108,000 tokens per second) with 100% token parity and no quality loss — a process that would normally take months. See the upstream pull request (covering optimizations from attention tuning to speculative decoding) for the specific changes. Argon also helped the team cut CPU usage for the YouTube comments frontend in half within Antigravity.
By tracing production telemetry, Argon identified that redundant emoji rendering alone accounted for nearly 45% of server load, prompting 12 targeted optimizations. These updates reduced server comment-load latency by 30 milliseconds and cut client-side end-to-end latency by 20 milliseconds.
AI Ultra is also required to access Argon in AI Studio and Android Studio (Canary). Argon is available in the Gemini API at a promotional price of US$2 (approx. HK$15) per million input tokens and US$10 (approx. HK$78) per million output tokens until November 30, 2026. After that, pricing will be US$4 (approx. HK$31) per million input tokens and US$20 (approx. HK$156) per million output tokens.
In the scientific community, Google is offering 10,000 scientists six months of free access to the “Google Antigravity highest usage tier.” Applications are now open.