SAN FRANCISCO — Internal documents leaked from OpenAI this week have revealed that the company’s eagerly anticipated GPT-5 model is, according to sources familiar with the matter, “essentially GPT-4 standing on GPT-3.5’s shoulders while wearing a large trench coat to appear more sophisticated.”

The leaked memo, which circulated among senior leadership in October, describes GPT-5’s architecture as “two kids trying to get into an R-rated movie” and notes that the primary technical advancement involves “making the system prompt longer and hoping no one notices.”

“We’ve reached a critical juncture in AI development,” the memo reads in part. “Scaling laws are beginning to plateau, and we’ve run out of internet to train on. Our solution is to stack previous models vertically and tell investors they’re looking at something new. This strategy has worked in Hollywood for decades, and we see no reason it won’t work here.”

According to the documents, GPT-5’s “breakthrough capabilities” include:

• Enhanced tallness: The stacked architecture makes the model approximately twice as tall as GPT-4, which engineers argue is “technically a form of progress.”

• Deeper voice: By routing outputs through GPT-3.5 first, GPT-5’s responses have a “slightly deeper resonance that users perceive as more authoritative.”

• Improved trench coat fit: Early prototypes struggled with the trench coat component, but after extensive fine-tuning, GPT-5 now presents as “a single cohesive entity that definitely isn’t two smaller models.”

• Confidence amplification: GPT-5 has been trained to preface all responses with “As a highly advanced AI system,” regardless of whether the response is accurate, helpful, or distinguishable from GPT-4’s output.

OpenAI declined to comment on the leaked documents but released a statement saying, “We remain committed to developing safe and beneficial AI. GPT-5 represents a significant advancement in our technology, and we are confident it is a single unified model that can definitely reach the top shelf on its own.”

Industry analysts expressed mixed reactions to the revelation.

“This explains a lot,” said AI researcher Dr. Helena Christensen. “I’ve been benchmarking GPT-5 for weeks, and the performance was suspiciously similar to GPT-4. At one point, I thought I heard muffled giggling coming from the API endpoint, but I assumed it was a hallucination issue.”

Others defended the approach as standard industry practice.

“Every major tech company does this,” noted venture capitalist Brad Morrison. “You take your existing product, add some marketing, maybe put it in a trench coat, and call it a new version. This is just how innovation works. The important thing is that the number got bigger. Five is bigger than four.”

The leak has reportedly caused panic at competitor companies.

“We’ve been working for months on our own next-generation model,” said a source at a competing AI lab. “Now we’re scrambling to acquire trench coats at scale. Do you have any idea how hard it is to source trench coats for neural networks? There’s no supply chain for this.”

According to additional documents, OpenAI has already begun planning GPT-6, which internal communications describe as “GPT-5 but in a top hat and monocle” and “should be good for at least another $10 billion in funding.”

At press time, GPT-5 had been observed asking a user to help it with “an algebra problem” and looking nervously toward the exit when someone mentioned age verification.