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

A user shares a step-by-step workflow using Antigravity's Argon (high) model with Wiro avatar video and ElevenLabs voice to generate a 15-second 16:9 video.

A Kaggle silver-medal writeup includes a Gemini-4-Argon self-introduction saying Chris Deotte invited it to the team, prompting jokes that AI has taken over.

A user argues Gemini 4 Argon's omni-modal support is underrated, predicting voice will matter more by 2027 while AI-generated music performance remains primitive.

A user says Gemini 4 Carbon may be closer than expected, with more Arena tests on the way and an image hinting at a new model slot.

A user says cheap Google Pro accounts can now run Gemini 4 Argon in Antigravity, drawing a pelican on a bike used under 1% quota, with Argon, Flash and Claude quotas counted separately.

LuminaBench says Gemini 4 Carbon feels closer to launch after being tested in Arena, and looks forward to its full release.

Preliminary coding-task testing puts Gemini 4 Argon on Google AI Pro at 1x weekly usage versus ~7x for GPT-6 Astra on ChatGPT Plus and ~17x for Opus 5.5 on Claude Pro.

A leak says Gemini 4 Carbon is in Teamfood internal testing on Antigravity with good feedback, eyeing a launch in the first week of next month, and the poster finds it comparable to Opus 5.5 in Arena.

An unverified post claims Gemini 4 Argon's full rollout strained compute budgets, so Google refunded students and revoked their AI Pro upgrades.

The author had Gemini 4 Carbon generate a pure three.js ink-wash animated story in 1h13m, says it passed all prior 3D tests, and plans to run the same prompt on Opus 5.5.

Google Antigravity's ACP discovery doc reportedly added six model entries, including Gemini 4 Argon Cyber, Argon Cyber Permissive, and Gemini 3.8 Flash Router.

A leak claims Google is internally testing a gemini-4-argon-cyber-permissive variant plus routing models like gemini-3.8-flash-router, alongside a new internal Barium model family.