Early Gemini 4 tester: beats FABLE 5.2 on UI tasks but token efficiency and reasoning need work

YashasGunderia · x · 2026-11-16

Responding to Google's call for Gemini 4 (Argon/Carbon/Hydrogen) feedback, a tester reports being impressed by its UI skills—iterating against a reference video until it matched perfectly, beating FABLE 5.2 on UI—and its creativity: last week it helped turn his laptop into a touchscreen, devising hand/eye/mouth cursor-control techniques. He flags token efficiency and the Antigravity harness as areas needing improvement.

Original post →