Allow models to "work together"

[davidwhearn]

For those who aren’t price-sensitive, it would be incredible to harness the power of OpenRouter with RooCode and enable multiple models to collaborate on a task simultaneously. For instance, o3 and Gemini 2.5 could jointly develop their optimal solution to a problem, then exchange it for feedback. They would iterate and refine based on each other’s insights, ultimately leading to a model implementing the code. This code could then be passed to another model for verification.

Of course, this would involve two or more LLMs conversing with one another, which could result in significantly increased token usage. However, it could be a remarkable achievement in amplifying the intelligence of the models.

Imagine a true Pilot and Copilot relationship! ;)

[BradKML]

This is similar to what some might consider an "LLM Router" when dealing with model switching (a task classification challenge), and a multi-agent framework when having models to wear different hats (a prompting challenge). There should be cursor rules or MCPs that does the job, but I would have to look into this first