The model selector has been removed in favor of agent mode, which automatically decides which models to use for different parts of the generation. Our tests show significantly better results using a separate model for reasoning than for actually outputting code, even within the same message response

Actually, it was better before. Now it only makes horrible user interfaces that look like they were made with Paint.

+1 I genuinely hope there's a solid rationale behind this decision because, as it stands, it feels somewhat arbitrary. In my opinion, v0 went from being the best frontend-generation model available to effectively unusable, clearly marking a step back. Many users—including myself—would gladly pay more for higher-quality models, so it’s frustrating that in order to use the tooling available in v0 we are restricted to using inferior models

+1 Same exact experience here, just today I spent around $5 on a few prompts that generated very little value for me. The previous version was fantastic, especially with the ability to chose the model.