(Credit: Zooey Liao/PCMag Composite; GettyImages/ OpenAI)
In the past, a new ChatGPT model meant a generational leap for AI chatbots. But GPT-5 doesn't appear to live up to that expectation. Don't get me wrong. I like the features and improvements GPT-5 promises: the ability to build apps from scratch, fewer hallucinations, more useful integrations, and voice chat upgrades. But none treads new ground. In a sense, ChatGPT-5 is like a new iPhone. It's not nearly as revolutionary as it was when it first came out, and its iterative improvements are becoming increasingly tougher to appreciate. And, unfortunately for OpenAI, I don’t see this changing until the company develops artificial general intelligence (AGI) or something very close to it.
What’s Actually New With GPT-5?
GPT-5’s headline upgrades are largely related to performance: You can expect fewer deceptions and hallucinations. And thanks to it combining reasoning models (such as the o-series) with faster, conversational models (such as the 4-series), you can expect more accurate responses, too.
Beyond that, though, not much stood out from the announcement livestream. Sure, you can create apps from scratch with GPT-5 (something Claude already does) and connect to various Google apps (something Gemini can do to a much greater extent), but these features, among others, are merely iterative.
If you are a programmer, you might be excited to see how GPT-5 will improve your coding work. But, for the rest of us, the new model doesn't bring any notable changes to ChatGPT's deep research, image and video generation, or internet search capabilities. Those kinds of major upgrades are what made past models exciting and special. I haven't yet had the chance to test GPT-5 and verify that there aren't any vast improvements, but I feel like they would have been a focus of OpenAI's presentation if there were any.
Why Isn’t GPT-5 More Impressive?
GPT-5 doesn't seem as impressive as past models precisely because ChatGPT is so impressive already: It can pretty much handle any task you throw at it. Need help with your homework? No problem. Want to write an email? Sure. Trying to debug a trick piece of code? Easy-breezy. Feel like reading a lengthy report on any subject you can imagine? Can do.
Of course, ChatGPT gets things wrong all the time, which is why there is room for improvement. But aside from making fewer mistakes and responding faster, even new features are tough to get that excited about. Take ChatGPT’s last few new, unique features, for example: Agent, Shopping, and Study Mode. How compelling is it to automate simple tasks that don’t take much time to do? How useful is it to shop with ChatGPT? And how great is studying with Study Mode versus asking ChatGPT questions in a regular chat? The answer is, in most cases, not very.
Apple had the same problem with the iPhone. The device ushered in the smartphone revolution, and for years, each new iPhone (and new version of iOS) upended the category in major ways. But as time went on, there was increasingly less to care about. We went from monumental upgrades that made it possible to stream high-quality video and play games anywhere to developments like a 20% increase in processing power.
AGI (or Something Close to It) Is the Next Milestone for ChatGPT
Although GPT-5 itself isn’t that exciting, AI, as a field, is still rapidly advancing. The ceiling for how many useful integrations you can add to a chatbot might not be that high, but these are all just stepping stones to AGI or at least its immediate precursor. That would amount to an exciting achievement that truly changes the way people interact with technology (for better or worse).
AGI, for all intents and purposes, is the kind of AI in science fiction: A machine that’s truly capable of learning and thinking by itself, not a complicated algorithm designed to take in prompts and spit out responses. GPT-6 doesn’t have to fully crack the code on man-made consciousness, however, because even just approaching that point will be revolutionary.
In the here and now, even with GPT-5, ChatGPT is still just an AI chatbot. Talking to it can approximate talking to a human, but nothing more. It will still get things wrong, give you responses that start to feel repetitive, and make images with strange errors and distortions. Take one look at social media, and you’ll find countless folks who can spot AI content from a mile away. But when that’s no longer the case and AI is operating autonomously? That’s a much, much bigger deal than the iPhone launch—or any other technology development for that matter.