GPT-6 Sol vs Opus 5.5 on a "Trojan Horse comes alive" video prompt
Arena compared both models on the same prompt: the Opus 5.5 version spins a black-figure vase like a real object, with day turning to night, the horse rolling through the gates, soldiers climbing out and Troy catching fire.
THEY PUT A 3,000 YEAR OLD VASE ON A TURNTABLE AND IT STARTED TELLING THE ILIAD BY ITSELF Arena dropped a side by side of GPT-6 Sol by @OpenAI vs Claude Opus 5.5 by @AnthropicAI on the prompt “The Trojan Horse comes alive.” Left one is a clean drawing. Nice vase, horse on wheels, little soldiers, done. Right one is a physical object sitting on a pedestal. The camera orbits it. The painted scene keeps going. Day turns to night. The horse rolls through the gates. Guys climb out of its belly. Odysseus is labeled in Greek. Troy catches fire. An owl shows up like it knows something. Sparks float off the clay. I showed it to a friend who actually studied this stuff. He went quiet for like twenty seconds then said that’s how a real black-figure amphora would move if it could. Then he asked if they scanned a museum piece. They didn’t. It’s generated. Scores for @claudeai Opus 5.5 aren’t even up yet. Arena runs on real tasks people actually submit, not lab puzzles. Go look at the right vase. Then try to go back to normal images.