I’m writing my longest-ever article (already at 7,500 words, as a warning). It’s both a review and an exploratory analysis of GPT-5 (what we know about it and, more interestingly, what we don’t know about it). I want to make it the most comprehensive piece on the topic to give you, in one place, all the context you need to be ready before the model comes out sometime this year.
You may be thinking: “How can anyone possibly put out a mini-book-size blog post on an AI model that isn’t out yet?” Fair question. The short answer is that it’s not just about GPT-5. The long answer is, well, you may want to read the article itself.
I’ll share and explain all the available information, the official bits, the speculations, and the leaks. I’ve also added my analysis and interpretations beyond what’s known (which I not-so-humbly believe to be the most interesting part). My research suggests GPT-5 will be more than a model: A paradigm shift. And, at the very least, the first of a new generation of AI. In line with this, what I bring you today is an excerpt from the article, a section I’ve titled “The GPT-5 class of models,” focused on Google and Anthropic’s reactions to GPT-5 and whether they’ll be able to respond or even match OpenAI’s big release.
To not break expectations, let me clarify that I’m not leaking specifics about GPT-5 beyond the hints we’ve already been served (which you may not be aware of anyway so it may be worth reading just for that). GPT-5 is OpenAI’s most jealously guarded secret, which they’ve been cooking for a long time in the opacity that characterizes them, so it’s hard to narrow down the options to determine what’s GPT-5.
The value of the article lies in the overarching analysis. Here’s a rough summary of what I’ll cover: meta stuff about GPT-5 (like the section I’m publishing today), the info we have (release dates and abilities), a clear picture of how OpenAI’s goals influence what GPT-5 can be, the relationship between the model and the scaling laws (size, data, and compute), the state of affairs at both the business and research levels, and the most speculative part (my favorite): Algorithmic breakthroughs (multimodality, reasoning, personalization, reliability, autonomy, etc.).
I hope to publish the entire thing soon (lest OpenAI decides to speed up the release given the constant movement in the industry, including from OpenAI itself). I’m off before I spoil the surprise. Here’s the excerpt as a sample before the big meal.