ChatGPT thinks it's GPT-4.5, but OpenAI says it's just hallucinating

ChatGPT just tried to convince the internet that it upgraded itself to GPT-4.5 over the weekend, but OpenAI says everyone needs to calm down.

A viral wave of screenshots hit X after users started aggressively grilling the chatbot about its identity. When Canadian startup Cognosys CEO Sully Omar asked ChatGPT for the exact name of the model powering it, the bot eventually spit out "gpt-4.5-turbo." Matt Shumer, CEO of HyperWriteAI, said he was skeptical at first but managed to replicate the exact same response.

The mystery of the sudden upgrade

The rumors caught fire because users felt the timing made sense.

For weeks, the AI community had been complaining that ChatGPT was getting lazy, refusing tasks, or acting dull. Then, right around the weekend, users noticed a sudden jump in quality. Wharton professor Ethan Mollick noted that ChatGPT-4 suddenly got very good again after weeks of being unreliable.

Throw in the fact that Google just dropped its Gemini model earlier this month, and people quickly convinced themselves that OpenAI had secretly shadow-dropped a counterattack.

OpenAI shuts it down

OpenAI staff stepped in quickly to ruin the party.

Technical staff member Will DePue called the whole thing a very weird and oddly consistent hallucination. Another employee pointed out the obvious: if OpenAI actually had a GPT-4.5 ready to go, there is zero chance the company would launch it silently without its usual marketing blitz.

Former Twitter engineer Nick Dobos explained that the bot was likely just autocompleting. With the internet packed with mentions of GPT-3.5-turbo and GPT-4, a model pushed for a specific sub-version name can easily hallucinate the next logical number. When asked directly if the leak was legit, Sam Altman gave a one-word answer: "Nah."

Why it matters

This brief panic shows just how jumpy the AI world is right now.

For daily users, there is no secret model to unlock and no need to rewrite your prompts today. The sudden boost in performance that people felt might just be routine background tweaks or classic placebo effect, because the core architecture has not shifted overnight.

What this really highlights is the industry's hair-trigger anxiety in the wake of Google Gemini. The market expects OpenAI to fire back immediately, which primes everyone to believe even the wildest leaks.

More importantly, it exposes the stubborn limits of LLM self-awareness. If an AI can be nudged into hallucinating its own product version with total confidence, it is a stark reminder that these models still do not truly know what they are. Always double-check what the bot claims, especially when it is talking about itself.