Gemini 4 just launched, and the frontier model race is now four competitors: Gemini 4, Opus 5.5, GPT-6 Astra, and Muse Wonder. Peter Yang ran all four through five concrete tasks: 3D model generation, game creation, animated shorts, website design, and writing. The opening benchmark alone, building Himeji Castle in code, tells you more about raw capability gaps than any benchmark leaderboard.
The value here is not the final rankings. It is watching four state-of-the-art models fail and succeed at the same prompts in real time. That granular, side-by-side comparison across creative and technical domains is harder to find than a press release and more useful than a spec sheet.
If you work in any field where AI output quality directly affects your time or deliverables, this test is a practical calibration tool. Watch it to decide which model deserves your subscription, not to confirm what you already believe.
[WATCH ON YOUTUBE →]