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

A screenshot post shows Gemini 4 Argon giving a conspiracy-theory-style answer, a user's hands-on jab at Argon's behavior.

Kimi K3.1, GPT-6 Astra, Gemini 4 Argon and Opus 5.5 each build the Eiffel Tower from one prompt; Gemini gets soft clouds but a stockier tower, and price doesn't buy accuracy.

A single-prompt 3D pelican wheelie benchmark pits Gemini 4 Argon (5m 12s) against Kimi K3.1 Pro (21m 05s, $0.56).

A designer built an app for rewriting plots and merging stories, using Gemini 3.8 Flash TTS, Nano Banana 2 Lite and Argon in Antigravity.

The poster introduces Alex (AWS)'s theory rebuilding physics from quantum information and had Gemini 4 Argon produce an infographic so non-physicists can grasp it.

A French post runs one close-up prompt of a weary man on a mountain summit through Gemini Argon and Grok Imagine 4.7, asking which captured a human gaze.

Gemini 4 Carbon is said to have left Arena, while Antigravity release files list gemini-4-carbon variants and a gemini-4-argon-max-mode-tf ID, suggesting a launch is near.

A post cites LisanBench's re-ranking without the edit-distance rule, with GPT 6 Astra medium first, Gemini 4 Argon high second and GPT 6.1 Sol medium third, noting all three confirmed loopers.

The author shares a full-codebase security audit prompt, recommends a top reasoning model, and says Gemini Argon 4 is notably sharp at it.

The poster says hands-on time with Gemini 4 Argon left them skeptical of Google's frontier chances and of Carbon's claimed capabilities.

A user argues Argon is the world's best model for writing, math and research, while coding is not top-tier but close to 6.1 sol.

A user pits Gemini 4 Carbon against Opus 5.5 on the same 3D airplane prompt, praising both for detail and lighting and asking viewers to pick a winner.