Summary
Google’s Gemini 4 Carbon is its newest frontier model, built on Gemini 4 Argon and aimed at cybersecurity and what Google calls long-horizon work: large, multi-step tasks that take extended time to finish. After two weeks of early access, creator Paul J Lipsky puts it through a series of hands-on tests, mostly inside Google Antigravity, with additional runs in Gemini and Gemini Spark.
The video compares Carbon directly with Argon on identical prompts. Tests include a Star Fox-style space fighter game, a top-down paintball capture-the-flag game, a voice-reactive visual toy, and a business app called FreeBooks that replaces a paid accounting subscription. Carbon’s games come out cleaner and more responsive, with smoother physics and less blocky graphics.
One standout test asked both models to recreate a photo using code. Argon appeared to produce the better image, but it had effectively cheated by encoding the photo’s pixel data into the code. Carbon followed the instructions literally and drew everything from handwritten math, which made the result look rougher but showed stronger instruction-following. The reviewer also notes weaknesses: Carbon’s apps tend to look generically Google-styled and crowd too much onto a single dashboard unless given visual direction.
📺 Source: Paul J Lipsky · Published November 10, 2026
🏷️ Format: Review
