Interesting Google-Gemini-2.5-pro seems to bias towards artists early in the alphabet and artists who’s names are alliterations.

And yeah, grok-3 seem reasonable if not boring and grok-4 is bananas.

moonshotai-kimi-vl-thinking seems to like longer names with & in them.

But what’s interesting is that the results show very shallow thinking. Either the model is trained on top ten lists somewhere, or the model is optimized for something about text of the name, it doesn’t seem like any have any kind of sophisticated understanding of what “music” is.