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

On the same one-shot task, Carbon finished in about 4 minutes in Arena with richer interactions, while Argon took about 10 minutes in Antigravity, used ~2% of weekly quota, and output cleaner, more minimal results.

The poster says Gemini 4 Flash and Carbon are about to drop, and criticizes Argon for trailing Astra, Sol 6.1, Opus 5.5 and Muse Wonder while still having the worst rate limits despite lower per-task cost.

A reposted video is presented as a demo of a new Gemini 4 checkpoint, referred to as Carbon or Argon.

In a joke-punchline ranking, LLMs still rate Astra funniest, with Gemini 4 Argon behind Opus 5.5, Fable 5.1 and Meta Wonder.

The author rates Kimi K3.1 around a 6 and says it is clearly worse than Gemini 4 Carbon.

A Portuguese-language joke says the company is automating everything with Jev and Gemini Argon, and devs are meeting to fire 20 CEOs.

A Chinese user says Kimi K3.1 Pro's 200-yuan plan burned 82% of a five-hour quota on a fifth of one video, joking it feels like a China edition of Gemini 4 Argon.

A six-step playbook for using Gemini 4 Argon to generate a Shopify 3D product customizer, then print and ship orders with claimed 40–70% margins.

A user tests Gemini 4 Carbon generating an ethereal cloudy cat chasing a butterfly in three.js within 57 minutes, praising its 3D and spatial reasoning while noting noisy audio and no further outputs afterward.

Wiz's PHP RCE blog credits Gemini 4 Argon for vulnerability discovery, exploit development, attack detection and patch validation, with Google providing mitigation guidance.

The poster says Gemini 4 Argon turned one Nano Banana image into a full gothic cathedral 3D scene, iterating geometry, textures and lighting through Blender MCP.

A report claims Google is building a System One decision model with a new createDecision method in the Gemini API, possibly powered by Gemini 4 Flash.