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.
Sure, Gemma 4 is good, but the Chinese are leading the way. Why? What is it about their incentives that led to this strategic decision? Even Mistral in Europe is more dedicated to open weights than any US lab.