I’ve been getting the same error. Very frustrating. Why not just allow explicitly adding providers (i.e., provider name + url + key) and then use the /v1/models endpoint to auto-populate the models for that custom provider? You could then have a prefix for each provider in the dropdown when selecting the model (and you could just leave the cursor models un-prefixed to keep the default experience the same).
Otherwise, you have to manually uncheck every model just to get the OpenAI key to verify (since it uses the first model in the list to verify).
Also related to this is that you can no-longer select a model and have “thinking” unchecked. When you uncheck “thinking”, it automatically toggles “auto-select” on.
