[andai]

Yeah that's a great point. In OpenRouter's model fusion article found that just running Opus twice (and combining the results) gave similar results to one Fable on a research task, because there is so much variability in what it does every time you run it, and web research tasks benefit from that variability.

(Unfortunately it also took twice as long and cost twice as much as Fable, but that's besides the point :)