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

An investor blogger claims Gemini 4.0 Argon ranks first worldwide in data analysis and reasoning, beating Fable 5.1 and Astra 6, without citing a benchmark source.

A Korean user reports that Gemini 4 Argon suddenly showed up in their account, calling it Google's first frontier model in a long time, and plans to test it.

MTSlive gives models simple prompts like 'make an animated short that makes me feel something' with a small fixed budget and a set of tools, and shows Gemini 4 Argon's output on that benchmark.

A Japanese user confirms that the Gemini 4 Argon model is available only on the Google AI Pro plan and above.

A user says they used Gemini 4 Argon to build a free image-to-3D tool so they could 3D print their cat and take it everywhere, with photos in the comments.

A Japanese user found a Google Workspace help page titled about using Gemini 4, but the model list has no 4 Argon and instead adds Gemini 3.8 Thinking.

Google's Alexander Chen shares a close-up of the stretchy drum machine demo by @kaloyaan from the new Gemini 4 Argon video, best watched with sound on.

The author shows a demo of Chronos-omega, a complex mechanical structure made with Gemini Argon and three.js.

A user says that giving Argon a single image and prompt produced a 30-second video with sound effects, and calls the result impressive and worth the wait.

An official demo shows Gemini 4 Argon and multi-agent workflows in Google Antigravity cutting a 100+ page translation calibration job that would take weeks by hand down to two days across 17 languages.

scaling01 posts an image warning that Gemini 4 Argon may be a "dirty looper," suggesting its Arena results could come from repeated prompts.

Noting that Anthropic's $20 plan includes Opus 5.5 in Claude Code, the author guesses Google Pro subscribers will soon get some Argon access in Antigravity and complains that redeeming AI Pro monthly credits is cumbersome.