Post #3232751

I was trying to figure out how good/accurate an LLM is for particular tasks vs. how much it costs to use, under the assumption that many loops with a mediocre model will yield better results than just a couple of loops with a super fancy one. I ended up with this: sweet is super smart and cheap, bitter is stupid and expensive; it kinda creates a 3d topographic map. A mountain of sweetness and the valley of bitterness, and we should all use Qwen3.6 Plus for most of our stuff