Hi, I have a custom prompt asking Cursor to identify its model version before starting a task. Although I have Sonnet 4.6 selected in the UI, the model keeps identifying itself as Sonnet 4.5 in the chat output. All other models identify correctly, but 4.6 seems to have this discrepancy. Is this a routing issue or just a model hallucination? I’ve attached a screenshot for reference.
Hey, this is a model hallucination, not a routing issue.
Claude and other LLMs don’t have exact info about their own version. They’re trained on older data and can’t reliably identify themselves. Cursor picks and sends the request to the correct model 4.6, but the model can call itself anything depending on the prompt.
The source of truth is the Cursor UI. If Sonnet 4.6 is selected there, then that’s what you’re using, no matter what the model says about itself in the reply.
I wouldn’t recommend relying on self identification in prompts. It’s not a reliable method.
Thank you 