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

The author tested a Gemini 4 Carbon checkpoint labeled gemini-4-argon in Arena, generating an interactive 3D crystal citrus in about 20 minutes and judging its 3D capability better than Argon.

Google releases Gemini 4 Argon worldwide, reporting over 35% more agent-based code submissions via Antigravity and 87% on the ERQA embodied reasoning test.

Pichai says Gemini 4 Argon burned nearly 25 trillion tokens in its first 72 hours, 3x Gemini 3.0, while Google rents ~110K Nvidia GPUs from SpaceX for $920M a month.

A user posts the classic pelican-riding-a-bicycle SVG generated by Gemini 4 Carbon, another Gemini 4 model name alongside Argon.

Greg Speicher calls a take that judges Google by Argon superficial and biased, citing Search, YouTube, Cloud, Waymo and TPUs.

A reply notes Google also claims Gemini 4 Argon helped find the vulnerability, with everyone eager to take credit.

Researchers using GPT-5.6 Sol plus Gemini 4 Argon and human review found a PHP flaw letting unauthenticated attackers run code or read secrets.

Google says its Gemini 4 Argon model helped discover and mitigate critical security flaws in PHP, per multiple news accounts.

After Gemini 4 Argon's wider release, Google's CEO discusses early reception, where the model still needs work, DeepMind leadership changes, compute spending, and Gemini's personal agent plans.

PHP 8.2–8.5 patched two critical unauthenticated RCE and memory-leak flaws (CVE-2026-109618/109619), with Wiz saying Gemini 4 Argon and GPT-5.6 Sol strongly assisted the discovery.

A user had Gemini 4 Carbon generate a three.js mechanical flower in about 1h43m, including an unprompted bee, and calls it a clear improvement over Argon.

DeepMind had Argon watch a human play cornhole and produce MuJoCo simulation motions, letting Apptronik's Apollo sink seven straight real-world shots.