Photo-Illustration: Intelligencer; Photo: Getty Images
The release of ChatGPT in late 2022 is commonly cited as a tipping point for the current AI boom, but if you were paying close attention at the time, it was an underlying model release — the upgrade to GPT-4 in early 2023 — that really captured the industry’s imagination, turning a fascinating demo product into something that people could actually use in the space of a few months. The implied rate of progress was dizzying, and rumors of an imminent GPT-5 started spreading almost immediately.
GPT-5 wouldn’t come out for more than two years, and now it has. Lots happened in the meantime, of course. OpenAI has a bunch of competitors, now, who are collectively propping up the entire American economy with AI investment. Sam Altman was forced out of his own company — over disputes about control, direction and “candor” — and then he returned. The AI race, such as it is, is now international, with major competition coming from China. Lofty debates about AI alignment have been effectively reduced to a conversation between AI firms and the Trump administration, which seems uninterested in regulation except when it concerns rooting out “wokeness.” ChatGPT now has more than half a billion regular users, who see it as a search-like answer engine, a productivity assistant, and, in some cases, an emotional companion. The company’s underlying models have been upgraded and diversified under about a dozen different names, and in the time GPT-5 has been in the works, the company — and industry around it — has pivoted to new techniques for improving models after the pursuit of pure scale started showing diminishing returns.
Anyway, GPT-5 is finally here, and so are reports from people who’ve been secretly testing it for a while, along with some breathless quotes from Sam Altman about how revolutionary it. “I think having something like GPT-5 would be pretty much unimaginable at any previous time in human history,” he said, of course. But how is it really? And what can it tell us about, you know, all this?
Development was harder than expected
Leading up to the release, multiple reports described a long and bumpy road to GPT-5, including a series of pivots as well as reports from testers that the model’s progress, while significant in many ways, wouldn’t blow people away. At The Information, Stephanie Palazzolo, Erin Woo, and Amir Efrati detailed some of the troubles:
The company’s business progress has masked some internal concerns about its ability to keep improving its AI and stay ahead of other well-capitalized rivals such as Google, Elon Musk’s xAI and Anthropic.
Problems had been brewing for months before the current year began. For much of the second half of 2024, OpenAI was developing a model known internally as Orion and intended to become GPT-5. According to people who worked on it, Orion was supposed to offer a big step up in performance compared to the current flagship, GPT-4o, released in May that year.
But the Orion effort failed to produce a better model, and the company instead released it as GPT-4.5 in February this year. It has since faded from relevance.
The company regrouped and shifted focus to its reasoning models, but even products based on those, The Information reports, “weren’t progressing the way OpenAI leaders and researchers had expected,” and didn’t seem ready as recently as June.
Testers like the model (for programming)
Programmer and prolific AI writer Simon Willison describes his experience testing GPT-5:
It’s my new favorite model. It’s still an LLM—it’s not a dramatic departure from what we’ve had before—but it rarely screws up and generally feels competent or occasionally impressive at the kinds of things I like to use models for.
Ethan Mollick, a business school professor and AI influencer, was more impressed:
GPT-5 just does stuff, often extraordinary stuff, sometimes weird stuff, sometimes very AI stuff, on its own. And that is what makes it so interesting.
The divide here is telling: the “kinds of things” Willison is talking about using models for coding and software development, while Mollick is impressed by GPT-5’s improved ability to build software without coding knowledge. Another AI developer zeroed in on a strange aspect of GPT-5, and perhaps of other models of the moment:
while gpt-5 is radically better at programming, it’s actually i think worse at most writing tasks! and that’s ok. i think that is the future. at least in the medium term, it will outsource these sort of tasks to another model. most programmers aren’t good at writing either.
— ben (@benhylak) August 7, 2025
Other ChatGPT users may be less impressed
Casual users who encounter GPT-5 through ChatGPT aren’t likely to feel like they’re using a completely different product — aside from a much simpler interface and a new “personality” feature in testing now — while people who use it for software development, or in a corporate context, are more likely to notice a major change.
OpenAI, like every other AI company, is all-in on coding assistance
ChatGPT is arguably still the only true, native, mass-market product of the current AI boom, a household name with a large and growing user base. But OpenAI’s competitors, most notably Anthropic and Google, have found success, and revenue, in coding-oriented products.
Much of OpenAI’s presentation on Thursday focused on the model’s coding performance, and GPT-5’s enterprise pricing was clearly intended to compete with cheap options from Google, which has been undercutting its competitors for the last year. The AI industry as a whole has shifted hard in this direction for two related reasons: One is that recent model generations have gotten much better at writing and debugging code; the other is that AI leaders have lately aligned on the slightly adjusted story that coding assistance is the actual path to building models that can improve themselves, launching AI into the realm of AGI, superintelligence, or beyond.
More practically, it’s an area where development has been fast and easy to measure, as opposed to many of the more diffuse sorts of tasks to which LLMs are being applied, or which they might theoretically disrupt — it’s also an industry with a ton of potential clients, many of whom can’t wait to lay off more staff. It was a focus of OpenAI’s messaging on Thursday, and appears to be an even greater focus of its model development, too.
The big AI firms are clustered together and moving as a group
A year ago — and certainly two years ago — consensus wisdom in the AI world would have been that GPT-5 would blow every other model out of the water, saturate a bunch of benchmarks, and set a new standard for everyone else to chase. Instead, intense competition and new scaling techniques have led to a different situation: A world in which companies are collectively spending hundreds of billions of dollars in a race that remains, at least in terms of measured capabilities, pretty close:
GPT-5 results on ARC-AGI 1 & 2!Top line:65.7% on ARC-AGI-19.9% on ARC-AGI-2 https://t.co/npAXRi1liu
— François Chollet (@fchollet) August 7, 2025
On Polymarket, which has a lively betting culture around AI, sentiment flipped after the release, from about 75 percent in favor of OpenAI having the best AI model by the end of August to an 81 percent chance that Google would return to the lead by then. The release of GPT-5, as impressive as some testers found it, was probably a relief for the other labs, whose current models — perhaps with the exception of Meta’s — are clearly on similar developmental trajectories as OpenAI’s. And for all the talk of superintelligence and PhDs in your pocket, some defensiveness came through in Altman’s messaging, between (AI-generated?) chart crimes and a fresh example of the now-classic “scientific error in the live demo” routine:
GPT-5 is the smartest model we’ve ever done, but the main thing we pushed for is real-world utility and mass accessibility/affordability.we can release much, much smarter models, and we will, but this is something a billion+ people will benefit from.(most of the world has…
— Sam Altman (@sama) August 7, 2025
OpenAI’s rivals are probably less relieved about GPT-5’s pricing: Free for basic access, and far cheaper for many enterprise and coding applications than some of its competitors. All these companies are already spending incredible amounts of money to win a race for new capabilities, far ahead of the revenues generated by related products. Now, these companies might be in a price war, too.
More screen time
Why Are There Betting Markets for Dildo-Throwing at WNBA Games?ChatGPT Is Now Encouraging Breaks But Not BreakupsSEO Is Dead. Say Hello to GEO.
From Intelligencer - Daily News, Politics, Business, and Tech via this RSS feed