I feel like ChatGPT router already acts differently (dunno since when). Responses to non existing models (edge cases?) are handled by the model or router model with a slightly different system prompt. * model name changed for testing purposes only

→ View original post on X — @testingcatalog