Inspired by OpenRouter's model fusion feature. There is a panel of models running in parallel and a judge to fuse the results.

I'm using it for my own deep research using tool calls. It's currently slower and more expensive for head-to-head prompts compared to OpenRouter's version, but with more work, it can be very similar.

Already, the responses I get are enhanced. Would love to hear what you think.