When you select a specific model, the request is always processed by that model without switching. So if you choose Claude 4 Sonnet, the response will come from that exact model.
Important: all Anthropic models are named “Claude,” and Claude 4 was trained on data that included information about Claude 3.5 Sonnet. Also, models accessed via API without a system prompt don’t have knowledge of themselves.
I’ve tested this multiple times, and the results are consistent.
Important note: According to Anthropic’s release notes, the model identifier claude-3-5-sonnet-20241022 that you’re seeing in the response is no longer available. This further confirms that the self-identification response is unreliable and outdated.