Creating a niche AI Benchmark with token anxiety (thijsbrits.nl)

Yo Can't help but see that LLMs are literally made to predict next tokens. Can you tell me how does this benchmark help my claude code? Or any other harness we use. It's quite unclear to me.