I noticed this recently with Opus 5.5: you can tell it to "make this more fun" for a game project and it... makes it more fun.
It's neat (if unsurprising) that Fable 5.2 has this ability as well.
My loose testing shows other models aren't quite as good at this yet. Here's the prompt "Build a mobile-friendly three.js game where a mouse has to hide from a cat" by Opus 5.5 Extra - https://claude.ai/artifact/RRyEndXLSPQbcig3BgzgCZ - and by GPT-6 Extra High: https://squeak-and-seek-mouse-1027.simonw.chatgpt.site/ - the Claude one is a better game, even though they look very similar at first glance.
Update: took it about 40 minutes but here's Fable 5.2 Extra's attempt at the same prompt: https://claude.ai/artifact/Agsqbng2MnkTPAU67it87o
I wrote about Opus 5.5's sense of fun in more detail here: https://simonwillison.net/2026/Oct/28/vibe-game-development/
It's crazy how even the intro text to the Fable 5.2 Extra's game still reeks of AI. Like, wow, the models can make a game that almost makes sense, but still can't write text that sounds remotely like a human was involved!