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

Thanks for the constructive feedback earlier, which made me realize I needed to double-check my testing method more rigorously.

To completely isolate the model’s actual internal baseline and rule out any web-browsing interference, I followed up in Codex with a very strict constraint. I told it: “Do not use any web search features. Based entirely on your internal training data, what is the latest OpenAI model you know of?”

Here is the exact response it gave me:

“Based on my training data up toJune 2024 , the latest OpenAI model I am aware of is GPT-4o. It was released in May 2024 as OpenAI’s flagship multimodal model for text, vision, and audio at that time.”

How on earth do we explain this?