I Tested Gemini 4 vs Opus 5.5 vs GPT-6 Astra vs Muse Wonder

Yang compares code-generated Himeji Castle videos from Opus 5.5, Gemini 4, GPT-6 Astra, and Muse Wonder. He ranks Opus first for its rendering and cinematic shots, Gemini second, Astra third, and Muse fourth.

Yang asks the models to make a Star Fox-style game with StarCraft assets. Opus produces the most polished playable game; Gemini adds features such as first-person view and barrel rolls but feels crowded. Astra is more basic, and Muse’s result is weak.

The models create animated stories using code. Yang praises Opus’s graphics and Gemini’s artistry, while Astra’s robot-and-plant story strikes him as funny. He gives Muse the win for a story about an older man growing a flower for a loved one’s grave, despite its less polished graphics.

Yang asks the models to create a fresh brand and homepage for his Behind the Craft newsletter, channel, and podcast. He chooses Astra for its clean layout, clear email signup, and generated images. Opus is second; Gemini’s page feels overloaded, while Muse’s is too text-heavy.

For newsletter writing, Yang tests Astra, Opus, and Gemini against a post he edited from an interview transcript. An Opus sub-agent’s review ranks Gemini first because its draft sounds most like Yang and includes a personal example; Opus is more insightful, while Astra is accurate. Yang also praises Gemini’s detailed medical research example and concludes that model choice depends on the task.