Welcome to Eye on AI! On this version...OpenAI releases GPT-5 and strikes a $1 authorities deal. AI homework helpers duke it out. Zoox will get a particular exemption.
One of many defining truths in regards to the world of generative AI is that even if you’re on high, the lead doesn’t final for lengthy.
And so, the 2 key questions popping out of OpenAI’s long-awaited launch of GPT-5 at this time are whether or not the brand new LLM may help the corporate reclaim the mantle of undisputed AI chief—and in that case, how lengthy can OpenAI preserve the lead?
OpenAI says GPT-5 delivers “extra correct solutions than any earlier reasoning mannequin,” and is “a lot smarter throughout the board,” mirrored by robust efficiency on educational and human-evaluated benchmarks. Its analysis weblog boasts of latest state-of-the-art efficiency throughout math, coding, and well being questions, and located that GPT-5 outperformed different OpenAI fashions throughout duties spanning over 40 occupations, together with legislation, logistics, gross sales, and engineering.
“GPT-5 actually looks like speaking to a PhD degree skilled in any subject,” OpenAI CEO Sam Altman instructed journalists in a pre-briefing on Wednesday. “One thing like GPT-5 could be just about unimaginable in some other time in historical past.”
Altman described GPT-5 as a “important step” alongside the trail to synthetic common intelligence (AGI), which, in accordance with OpenAI’s mission assertion, is outlined as “extremely autonomous methods that outperform people at most economically worthwhile work.”
It’s unclear whether or not this mix of velocity, energy and options will probably be sufficient, nevertheless. Some two years within the making (GPT-4 was launched in March 2023), GPT-5’s launch has taken longer than many trade insiders anticipated, as OpenAI has adjusted its method in response to trade adjustments. And whereas ChatGPT now boasts a powerful 700 million weekly customers, OpenAI has confronted rising stress over the previous yr as rivals poach its expertise and race forward on rising AI methods like long-context reasoning and autonomous instrument use. Along with Huge Tech opponents like Meta and Google, there’s a wave of startups based by OpenAI’s personal former researchers, together with Anthropic, Pondering Machines, and Protected Superintelligence. And naturally, there’s the brand new crop of highly effective Chinese language fashions, like DeepSeek, vying for international affect.
Whether or not GPT-5 propels OpenAI again to the highest of the AI hill will develop into clear within the days and weeks forward, as researchers put the mannequin by means of its paces, testing it towards the likes of different elite fashions, together with Anthropic’s newest Claude mannequin and Google’s Gemini.
OpenAI pushes to remain within the lead
With GPT-5 now lastly out, OpenAI CEO Sam Altman acknowledged that staying on the frontier means one factor: relentless scaling.
In AI, scaling refers to the concept that fashions get extra highly effective as you enhance the quantity of knowledge, computing energy, and mannequin elements used throughout coaching. It’s the underlying precept that drove progress from GPT-2 to GPT-3 to GPT-4—and now GPT-5. The catch is that every leap requires exponentially extra funding, significantly in AI infrastructure—for OpenAI, that features its Stargate Venture, a three way partnership it introduced in January with Softbank, Oracle and funding agency MGX with a objective to to speculate as much as $500 billion by 2029 in AI-specific information facilities throughout the U.S.
When requested whether or not scaling legal guidelines nonetheless maintain, Altman stated they “completely” do. He pointed to higher fashions, smarter architectures, higher-quality information, and considerably extra computing energy as the trail to “order-of-magnitude” enhancements nonetheless forward.
However that type of progress comes at a value. “It’s going to take an eyewatering quantity of compute,” he admitted. “However we intend to proceed doing it.”
That after all, requires huge sum of money and partnerships.
On the brilliant aspect, OpenAI has roughly doubled its income within the first seven months of 2025, hitting an annualized run price of $12 billion—up from about $6 billion at first of the yr, in accordance with a current report by The Data. That interprets to $1 billion in month-to-month income, fueled by surging demand for its ChatGPT merchandise throughout each client and enterprise markets. Weekly lively customers for ChatGPT have jumped to round 700 million, up from 500 million throughout all OpenAI merchandise as of late March. And earlier this week OpenAI launched a free, open-source mannequin—an uncommon transfer for an organization usually criticized for its closed method over the previous half-decade—suggesting confidence that its premium providing, which is now GPT-5, will proceed to dominate.
However there are some huge challenges forward, nevertheless. For one factor, the partnership between Microsoft and OpenAI—that started with a $1 billion funding in 2019—is getting into a extra fraught and complicated part. Whereas Microsoft has invested greater than $13 billion and retains unique rights to OpenAI’s fashions by means of Azure, tensions have emerged over income sharing, AGI management clauses, and overlapping product methods. And the Stargate challenge is reliant on partnerships with corporations like SoftBank, whose funding stipulations are tied to OpenAI’s nonetheless unresolved efforts to overtake its company construction.
There’s rather more to say about GPT-5. Journalists obtained the complete set of supplies from OpenAI, together with a analysis weblog, system card, and security card, a mere 90 minutes earlier than releasing the mannequin, so I nonetheless have rather more to undergo. Keep tuned for extra!
Additionally: In lower than a month, I will probably be headed to Park Metropolis, Utah, to take part in our annual Brainstorm Tech convention on the Montage Deer Valley! Area is proscribed, so should you’re curious about becoming a member of, register right here. I extremely suggest: There’s a unbelievable lineup of audio system, together with Ashley Kramer, chief income officer of OpenAI; John Furner, president and CEO of Walmart U.S.; Tony Xu, founder and CEO of DoorDash; and lots of, many extra!
With that, right here’s the remainder of the AI information.
Sharon Goldman
sharon.goldman@fortune.com
@sharongoldman
AI IN THE NEWS
OpenAI strikes a $1 authorities deal. OpenAI has struck a cope with the U.S. authorities to supply federal businesses entry to its frontier AI fashions—together with ChatGPT—for simply $1 over the following yr, in accordance with a joint announcement with the Normal Providers Administration (GSA). The partnership displays months of behind-the-scenes outreach by OpenAI CEO Sam Altman and his staff, who’ve been cultivating ties with the Trump administration since earlier than Donald Trump’s return to the White Home in January. In a press release to Wired, Altman framed the settlement as a part of President Trump’s AI Motion Plan, saying it could assist public servants leverage AI to higher serve the American individuals.
AI homework helpers duke it out. One week after OpenAI launched a ChatGPT Research Mode, Google has jumped into the AI tutor sport with a brand new Guided Studying mode. Google says it labored intently with studying consultants, academics, and college students to develop Guided Studying, which is built-in inside Gemini. Like ChatGPT’s Research Mode, Google’s Guided Studying is designed to encourage customers to work by means of issues by means of “probing and open-ended questions,” quite than simply spit out the solutions. With lessons at many faculties and universities set to start out within the coming weeks, the massive query will probably be whether or not college students truly need an AI examine companion, or are content material with an AI reply machine.
Zoox will get a steering wheel exemption. The evolution of the robotaxi took a step ahead Wednesday as U.S. federal regulators gave permission for Amazon’s Zoox to check self-driving shuttles with no steering wheel, pedals, or different handbook controls. The transfer clears up some extent of ambiguity within the self-driving automobile trade. Whereas Zoox had beforehand claimed its autos had been compliant with federal motorized vehicle requirements, the Nationwide Freeway Visitors Security Administration’s guidelines appeared to counsel that was not the case. The NHTSA’s transfer makes it clear that U.S.-built self-driving automobiles do certainly want handbook controls to function (even for testing) on public roads, however grants Zoox an exemption—and it opens the door for different robotaxi corporations akin to Tesla to get a waiver.
FORTUNE ON AI
AI is already upending the company org chart because it flattens the space between the C-suite and everybody else —by Beatrice Nolan
How Palantir—an organization too small to make the Fortune 500—turned one of many world’s 25 most precious corporations —by Jessica Matthews
OpenAI launches its first open mannequin in years so it may well cease being on the ‘fallacious aspect of historical past’—whereas nonetheless maintaining its most precious IP below wraps —by Sharon Goldman
AI CALENDAR
Sept. 8-10: Fortune Brainstorm Tech, Park Metropolis, Utah. Apply to attend right here.
Oct. 6-10: World AI Week, Amsterdam
Oct. 21-22: TedAI San Francisco. Apply to attend right here.
Dec. 2-7: NeurIPS, San Diego
Dec. 8-9: Fortune Brainstorm AI San Francisco. Apply to attend right here.
EYE ON AI NUMBERS
$364 billion
The overall sum of money that tech’s 4 high hyperscalers—Meta, Microsoft, Google, and Amazon—plan to spend on 2025 capital expenditures as they construct up their AI and cloud infrastructure. The businesses are pouring big sums of cash into constructing and increasing information facilities crammed with costly Nvidia GPUs in a race to remain forward in AI and to attain AGI.
