Comparison of 3D game demos across models

I love Meta, but I have some tough love. Opus 5.5 is absolutely mogging Wonder on every single 3D game demo we’ve seen, and I love Alexandr because he can admit that they didn’t explicitly train Wonder on 3D games/visual representations. I know he’s heard this, however, why wouldn’t you? Theo saying 3D demos don’t represent real coding is utterly ridiculous. If you work with models on 3D visual representations, you know that they’re testing geometry, coordinate transformations, camera systems, physics, rendering, and actual interactive environments. A model can’t just be RL’d on game demos and build unique, never before seen 3D representations without having genuinely good spatial reasoning and creativity within the model. What do you think game developers do all day? I’m saying this as someone who used to be one. Alexandr and the Meta team should have known that when people are scrolling on X, you’re not going to see posts saying, “Wow, Meta Wonder found me a house on Zillow that matches my exact preferences!” or “Oh wow, it one shotted this extremely economically useful spreadsheet!” Few people are going to repost and get excited about that. People LOVE demos because they make the model’s capabilities actually tangible and something we can use, play with, and watch every day. Watching an LLM generate an entire playable 3D world from a prompt and Reaching AAA game quality is an almost insurmountably difficult challenge for these models. I think Meta seriously underestimated how important that is. I don’t think people are pushing Meta Muse wonder as far as they could be so I want to give it a fair chance. The demo below is not a fair representation but shows how good Opus 5.5 can look when paired with unity.