ChatGPT-5: a small step for intelligence, a giant leap for normal users

I’m back from two weeks of summer holidays in North Macedonia, Bulgaria, and Italy; and I’m begrudgingly ready for the second half of the year... And what a week to be back! On Monday morning, you should have access to a different version of ChatGPT than you did at the end of your work week on Friday: ChatGPT-5. The reaction to the new model over the weekend has been mixed, but I agree wholeheartedly with Peter Wildeford's view here: ChatGPT-5 is a small step for intelligence, but a giant leap for the world's 700 million "normal" ChatGPT users (https://lnkd.in/dHyM8-fR). ChatGPT-5 does little for power users like me, but what most people will see today and in the weeks to come represents the democratisation of cutting-edge, much more powerful AI tools - tools that power users have been going *gaga* over for the last four months. Let me say more: I’m on the record (see link below) saying that I didn’t find the first generation of much smarter, so-called "reasoning models" very helpful. I barely used ChatGPT’s o1-level models from September 2024 onwards, despite repeatedly trying them for various use cases. They couldn’t access the internet (and so used outdated info), couldn’t take attachments or images for context, etc. I’m sure they were great for math and logic, but for my work and life, they were fundamentally unhelpful. However, in mid-April, OpenAI launched an upgraded reasoning model named o3, which added internet and tool use, multimodality, and attachments - and I was immediately obsessed. I have constantly run up against usage limits. I have a real second brain in my pocket, and it is truly 🤯. Sam Altman characterized the transition from GPT3 (2020), to GPT4 (2023), to GPT-5 (2025) as the transition from high school, to college, to PhD-level models. I broadly think that’s right. o3 was the first model that really helped me go much deeper even in areas in which I’m a bona fide PhD and real-world expert. It was also the first model that truly scared me regarding the future of humanity and the future of work. However, you had to know about it, and you had to manually select it in the "model picker" to use it. Even among paying ChatGPT users, only 7% of users used any of the reasoning models (sticking instead with the basic 2023-era models up to today). Of non-paying users, less than 1% used reasoning models. By far the biggest new thing that ChatGPT-5 has done is add a so-called "model router" that looks at your question and assesses whether to engage a fancy, smarter, reasoning model to answer, or whether to respond quickly with a smaller model for queries that don't need extra firepower. Much more user-friendly! I'm excited for everyone to catch up with two years' and billions of dollars of AI advances.