Bijan Bowen gives an initial assessment of Gemini 4 Hydrogen and runs a series of hands-on tasks, including a browser-style interface, games and a robot-arm exercise. He describes his judgments as subjective and notes that hardware performance and reasoning settings affect results.
Bowen tests a 3D-printable model created from a reference image and reports that it printed without supports. He also shows a game with failed voice assets and a skate game that improved after follow-up prompts but still had weaknesses.
Most tasks used an agent swarm featureA multi-agent system contains multiple AI agents that interact, coordinate, divide work, or influence one another while pursuing tasks. and all used the selected high reasoning setting, while the browser task did not use the swarm. The different prompts and settings limit direct comparisonsEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. with prior runs; usage and pricing observations reflect the presenter's account at recording time.
Watch on YouTube
