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

A prompt for Claude Code to inventory hardcoded model strings and centralize routing before Gemini 4 Argon's intro pricing ends Nov 30 and doubles on Dec 1.

A side-by-side video runs the same prompt through Gemini 4 Argon and Opus 5.5 with no edits or cherry-picking, asking viewers which molten-elephant result wins.

The author used Gemini 4 Argon to build a polished Minecraft-style demo in roughly 1 hour 20 minutes, with the model even choosing the trailer shots.

An indie developer shares gameplay from Ergod World, a browser MMO built with Three.js alongside Claude Opus 5.5, Codex/Astra/Sol and Gemini/Argon.

The author says they used Gemini 4 Argon with VS Code and Blender to make an 8-second 3D horror corridor clip that looks like a game trailer, and asks if readers want a tutorial.

A short Portuguese post complains that Gemini Argon performs no better than Gemini Pro.

In a browser marine lagoon build-off among Claude Opus 5.5, GPT-6.1 Sol, GPT-6 Astra and Gemini 4 Argon, Argon cost only $0.22 but was judged visually the worst.

An Arabic-speaking user says Gemini 4 Argon is strong inside the Gemini app but poor in coding environments, calling it useful only for analyzing student work.

A weekend project built with Gemini 4 Argon in AI Studio renders a falcon sharing the sky with starlings using only three.js and no other libraries.

A question aimed at developers asking which model they actually use: Gemini 4 Argon, Haiku 5.5 or GPT-6.

Raia Hadsell congratulates a team whose four Gemini 4 Argon agents wrote code and shared learnings for 27 hours to beat the Ender Dragon in Minecraft.

Given a real work bug and hints to the relevant files, Gemini 4 Argon scanned 30+ unrelated files and invented a nonexistent DB index column, while Opus 5.5 fixed the same task.