I haven’t successfully found a use case for Sol Ultra yet. I’ve run two days of side by side xhigh and ultra runs for planning and impl and I haven’t noticed any tangible quality change. I can see the difference in execution and token usage. Maybe I’m holding it wrong, but how?