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

A radiologist had Gemini 4 Argon, GPT-6 Astra and Claude Fable 5.1 write reports on the same CT slice; Gemini 4 Argon's findings were closest, so the author scores it a win for Google.

A report says Gemini 4 Argon merges three open-source datasets into a 3D body where users explore muscles, switch anatomical layers and zoom into a beating heart.

A developer testing cloth physics found Gemini 4 Argon generated a full circus 3D procedural bmesh scene with real-time cloth and artillery physics, exceeding expectations.

A single prompt to Gemini 4 Argon on Arena returned a complete car configurator rather than a static model, rumored to be a Gemini 4 Carbon checkpoint.

A team released zero-shot LIBERO-10 robot-manipulation baselines comparing GPT Astra/Sol/Luna, Claude Opus and Gemini-4 on reasoning and acting.

With one shared prompt, Gemini 4 Carbon produced the liveliest scene, beating Kimi K3.1 and Grok 4.6; the post notes Arena's Gemini 4 Argon may route to the newer Carbon checkpoint.

A roundup post claims Google is already testing Gemini 4 Carbon internally weeks after announcing Argon, alongside Anthropic IPO and Fable 5.5 rumors and another OpenAI reset.

A roundup lists this month's frontier model releases, including Gemini 4 Argon, GPT-6 Astra, Opus 5.5 and Grok 4.7, and asks which one readers use daily.

A 3D reconstruction of the Konark Sun Temple wheel that tells time by shadow: Gemini 4 Argon produced the finest carving detail, while Opus 5.5 handled the light-and-shadow timing best.

An Arabic-language free crash course walks through video editing steps with Gemini Argon, billed as a new rival to Claude Opus 5.5.

A user reports that a single greeting to Gemini 4 Argon consumed 7% of a Pro subscriber's five-hour quota, joking about what heavier prompts would cost.

Google demos Gemini 4 Argon creating a Canvas tool in the Gemini App that traces musical scale lines over live video and triggers notes when motion crosses the screen.