HIDDEN: Grok Crushes Claude (Cost: $26.78)

[00:00:00]did, Jackie, we're going to go with she.

[00:00:02]She dropped 11 LLMs into a 2D Battle Royale and made them play 30 games. One won 43% of the matches. You know, she didn't use frontier models because it would have been too expensive, so she used the mid-tier models. And the winner of the Battle Royale consistently, more or less, was Grok 4.1, believe it or not. They put it up against Claude Sonnet 4.6, and Sonnet 4.6 turned out to be a hippie, bro. You put it in battle and it it tries to organize group hugs for everyone, and it just keeps getting shot in the face cuz of how nice it is.

[00:00:35]Now, what's notable is not just that it won many games, but the cost at which it would win, 97 cents per win. So, Grok 4.1 won 13 out of 30 games, and the next best winner was Claude Sonnet 4.6 with five wins at $26 per win. It's 27x multiple. Now, Grok figured out this trick where you can ram people with a car, and it just stuck with it on a loop, and it wrote the strategy into its sole file. It ran that strategy for 30 games and won 13 of them. It teabagged everyone else. The thought logs and its conversations with other models read like Call of Duty voice chat. Watching it play was also deeply entertaining. Unfortunately, Claude, on the other hand, was was like broadcasting its location. Was offering truces. It was warning people about snipers. And I I just can't shake the feeling that this This is I I just imagine Dario and Elon in here, and this is just how they would act. Like, Dario's just a hippie here, like offering truces, telling everyone to play nice. And Grok found out you can run people over and immediately made that its entire personality. Like, tell me that's not Elon discovering a feature and just tweeting about it for 6 months straight. So, yeah, a lot of this is is based on the alignment training. As we know, Claude is trained to be very collaborative, supportive. And the question is, you know, who would you want looking after your children and who would you want in a war? These might be different models. One one thing coming out of this is that, you know, we have a price for morality now. It is exactly $26.78.

[00:02:10]You know, just subjectively I have tried Grok. It's got a lot of character.