Comment by pimeys

It thinks a bit more, but if the cache reads are free or almost free, and it blasts through 10x faster than GPT Sol/Terra/Luna, it doesn't really matter. Just don't use the max thinking for implementation, go with high or low.