There’s a lot to unpack so I’ll just get started…
I’ll start with what OpenRouter is, in case you haven’t heard of it. It is a platform that gives you access to a lot of AI models from one place (read: single API key). So, for example, instead of having access to ~10 models from ChatGPT, you can access HUNDREDS of models from OpenRouter.
It’s a pretty big step-up if you’ve only used chatbots.
Anyway, Fusion is OpenRouters new tool (still in beta but it is usable) that takes your prompt (question/instruction/request/etc) and sends it to multiple models. Then it takes the output from each model and sends that to a single “judge” model. The judge looks through the answers to give you the best combination. This approach can help a lot with accuracy and/or giving different perspectives.
In early testing, Fusion’s results were near or better than the top individual models (yes, even Fable 5).
It’s not all rainbows and unicorns though. There are, of course, downsides. For example, it will take longer (maybe a LOT longer). It will cost more since you have to pay for each model and the judge. You have to understand what is happening enough to select good models for the task(s). There are probably more downsides but those are the first that come to mind.
If you are still interested, there is a lot more to this.
Here is the documentation on Fusion on OpenRouter’s page - https://openrouter.ai/docs/guides/features/plugins/fusion
Or if you’re ready to try it out, here is the link to Fusion (you’ll need an account with OpenRouter and possibly credits if you are using paid models) - https://openrouter.ai/fusion
I think that this is a really good example of where things are heading and you will see a lot more tools like this coming out in the near future.
(But what do I know)