Stealth Model Swap? GPT-5.5 High Claims Knowledge Cutoff is June 2024

js_tear:

How on earth do we explain this?

The response is either a hallucination or, more likely, it’s using a system prompt where that is stated.

You cannot “ask” a model about which GPU it’s running on either btw

LLMs are token factories. They are not usually aware of their own configuration. This will be in the metadata of the model that is inaccessible during inference.

You should engineer assuming you can’t know the cut-off date. Then you can switch models later without relying on this.