One image - four AI models.
Which one did it best? I gave Grok 4.6 Build, ChatGPT-5.6 Sol, Claude Opus 5, and Gemini 3.7 Flash the exact same static image of a tropical beach and asked them to turn it into an interactive WebGL animation The results were completely different
This should straight up not be allowed
Moon dev: "trading is just math, so i built the exact same brain twice and let them fight to the death" "i put opus 5 against gpt 5.6 and one of them did something i genuinely was not expecting"
My point still stands. Without vision the model cant self fix.
Tencent Hy4 preview on the NYC prompt, run in opencode. Opus 5 underneath. To be fair it tested its own code and improved on the first round, it added more than it started with. But its all buildings. Ground half
3D candy shop sim today, fully built on Crayon Pro with Opus 5!
You have reached the end of the archive
All of Claude Opus 5