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

A user citing leaker Lyra says Gemini 4 Flash is coming, 3.8 Flash will be phased out, and Argon's separate quota is paving the way.

Developer mehulmpt reports that Gemini 4 Argon gave significant help in developing a PHP2Shell exploit.

On the same prompt, Carbon beats Argon on water reflections, dynamic weather, and voxel textures; it takes 2–3x longer but the visual leap justifies it.

JarrettYe says Gemini 4 Argon reproduces his Zhihu posts so well he jokes it can role-play him as Thoughts Memo.

Two CVSS 9.2 PHP core engine flaws were emergency-patched on Oct 21; Wiz Research says its Atlas system with Gemini 4 Argon and GPT-5.6 Sol helped locate memory conditions and adapt exploits across versions.

A user argues Gemini 4 Flash, not Argon, will be Google's real comeback, without offering benchmarks or hands-on evidence.

A short hands-on note saying Google's Gemini 4 Carbon almost solved the author's SVG generation test.

A user testing Gemini 4 Argon reports hitting the usage limit after just two prompts, complaining about how fast quota runs out.

An image post jokes that after building Gemini 4 Argon, billed as the world's best model, Google also dropped Codex inside just in case.

In a clip with Gavin Baker, Jensen Huang lists demand at OpenAI, Anthropic and Grok and notes Gemini 4 just launched, arguing compute demand can back Nvidia's chips.

An observer's read of X sentiment: Grok and Claude Opus 5.5 are well received, OpenAI is souring, and Gemini 4 Argon's release hasn't turned Google's mood around.

Wiz says Atlas, GPT-5.6-Sol and Gemini 4 Argon found and chained decade-old flaws like PHP2Shell in under a week, with Gemini 4 contributing notably.