Wow, these numbers are insanse! I tried it yesterday and it worked beautifully well. It also responded the way I wanted every time, I didn't have to spend time prompting it on how to respond properly (unlike Grok 4 expert, which tends to yap a lot), it just knew.

Todays models have gotten so good that at this point, whatever I run, just works and helps me in whatever. Maybe I should start noting down prompts that some models fails at.

Also, my experience with it wasn't that good; but it was looking good on benchmarks ..

It seems benchmark maxing, what you do when you're out of tricks?

Ohhh, so Qwen3 235B-A22B-2507 is still better?