Aggregating and synthesizing models, does it work?

I often see news debating which ai model is best or find myself switching between different models when one doesn't quite get the answer. For example, sometimes o1 isn't getting it so I go try with deepseek and it gets it, or vice-versa.

I even started copying outputs from one model to another, pasting and asking "What do you think of this response and thinking, could it help us solve the issue?"

That got me thinking: rather than focusing on finding the single "best" model, why not leverage multiple top models simultaneously and synthesize their combined insights?

I've been testing this approach with common "challenges" such as the how many "r" or "l" letters are in a word, or the generate falling balls in a rotating hexagon, as well as more difficult problems. The combined responses have consistently been excellent.

What do you think?