Video · Bijan Bowen ·
Gemini 4 Carbon Is INSANE – Seriously, This Model Is Ridiculous!
In short
The video hands-on tests Gemini 4 Carbon across browser workflows, C++ games, 3D scenes, agent swarms, robotics, and engineering tasks. It finds a major improvement over Argon, especially in 3D generation and agent-swarm output, making Carbon a strong option for building polished prototypes and solving practical design problems.
Key points
- Gemini 4 Carbon is described as a significant leap over Argon, particularly for 3D generation.
- The Browser OS test produced the strongest non-Claude result the creator had seen, with interactive games and tools.
- The C++ skateboard game improved substantially, gaining richer environments, physics, vehicles, and visual fidelity.
- The Jerry’s Apartment test delivered the most accurate floor-plan replication seen in these tests.
- The Subway FPS was better than Argon’s first result but remained relatively simple and showed performance lag.
- Agent swarm generated a polished pool-party game in under an hour and was identified as Carbon’s standout feature.
- Carbon helped diagnose a car problem and generate an OpenSCAD spacer model, showing practical engineering value.
Written by an AI model from the video’s captions and description.
In this video, we take a hands-on look at Gemini 4 Carbon, testing just how capable Google's newest model is across a range of difficult coding, creative, and engineering tasks.
In this story
A tool or model opens in the Heidelberg AI Curriculum: what it does, what it costs, how to start. A repo opens on GitHub.



