Gemini 4 Pro / Flash: benchmarks, hands-on tests and leaks

AndroidDev used Gemini 4 Argon to convert a basic 2D maps app into a 3D world, and points to Antigravity and Android Studio as ways to try it.

Google says Argon was tested by over 100,000 employees, supports 1M-token output, and improves writing, raw knowledge, and persistence in coding.

A user's screenshot indicates Gemini 4 Argon's training knowledge runs through March 2026, newer than the existing Gemini 3 series.

Gemini 4 Argon is available globally in the paid tier of the Gemini API and platform, in Antigravity for Ultra subscribers, and in the Gemini app and Search for Pro and Ultra subscribers.

Six Every staffers tested Gemini 4 for a week and found it excels at writing and greenfield apps, but refactoring succeeded only 21% of the time and Antigravity trails Claude Code by about six months.

Google announced Gemini 4 Argon, aimed at the hardest coding and knowledge work tasks, available to developers and enterprises on the paid Gemini API tier, to Ultra subscribers in Antigravity, and to Ultra and Pro subscribers in the Gemini app and Search.

A video post claims Google revived 10,000 businesses with Gemini Argon 4, a hyperbolic marketing-style take from launch day.

A short post announcing that Gemini 4 Argon is here, reflecting immediate launch-day reaction.

The new LiveBench 2.0 benchmark shows Argon 4.0 behind GPT 6.1 Sol and Fable Astra but ahead of open-source DeepSeek Flash and well ahead of Muse Wonder.

Manus says it has launched Gemini 4 Argon in partnership with Google Cloud, for tasks including coding, planning and 3D modeling.

A tester with two weeks of access says Argon is not just benchmark-tuned, matching SOTA in games, SVG and 3D at $2/$10 pricing, though Google's harness and sign-in experience hold it back.

9to5Google reports the Gemini app has begun rolling out Gemini 4 Argon to AI Pro and AI Ultra subscribers.