benchmarks that map to real use for Kimi

[vectorcrate98]

I figured I'd ask here before changing everything.

I'm trying to decide whether to keep the current stack or change course. Main question is how people here are handling prefix caching.

Would appreciate practical advice rather than theory.

[deltapoint17]

My first guess would be this: Before rebuilding anything, I'd test one clean baseline with Kimi and compare from there.