OpenAI unveiled a highly anticipated new version of its flagship ChatGPT on Thursday, highlighting “major” improvements in artificial intelligence as global competition in the field intensifies.
The company said ChatGPT-5 is now free for all, serving nearly 700 million weekly users.
Co-founder and CEO Sam Altman described the update as “undeniably a generally intelligent model.”
AGI Remains a Work in Progress
Altman acknowledged that the journey toward true artificial general intelligence (AGI)—machines that think like humans—remains unfinished.
“This model doesn’t continuously learn from new information as it’s deployed, which I think should be a hallmark of AGI,” he explained. “Still, the leap in capability is enormous.”
Industry experts say a new AI era could transform how people live and work.
“With AI advancing so quickly, the prospect of developing superintelligence is now within sight,” Meta CEO Mark Zuckerberg noted in a recent memo. “I believe this marks the start of a new era for humanity.”
Altman Predicts Massive Gains Ahead, Despite Costly Compute Demands
Altman added that there are “orders of magnitude more improvements” ahead on the road to AGI. “Of course,” he said, “achieving that demands staggering investment in computing power—but we fully intend to keep making it.”
Since ChatGPT’s debut in late 2022, Amazon, Google, Meta, Microsoft, and Musk’s xAI have invested billions in AI.
Earlier this year, Chinese startup DeepSeek disrupted the industry with a model that achieves strong performance while relying on more affordable chips.
InAmid fierce global competition, Altman said ChatGPT-5 now leads in coding, writing, healthcare, and more.
GPT-5 Feels Like Talking to a PhD-Level Expert
“GPT-3 felt like talking to a high schooler—you might get the right answer or something random,” Altman said. “GPT-4 was more like speaking with a college student. GPT-5 is the first time it truly feels like you’re talking to a PhD-level expert on any subject.”
He predicted that the ability to instantly create software—what he calls “vibe-coding”—will be a hallmark of the ChatGPT-5 era.
In a blog post, British AI specialist Simon Willison shared his impressions after gaining early access to ChatGPT-5.
“My take: it’s simply good at what it does,” Willison wrote. “It may not be a big leap over other models, but it’s highly competent—rarely wrong and often impressive.”
However, Musk claimed on X, formerly Twitter, that his Grok 4 Heavy AI model “outperforms” ChatGPT-5 in intelligence.
OpenAI Emphasizes Trust and Safety in ChatGPT-5 Design
OpenAI’s safety research lead, Alex Beutel, said ChatGPT-5 aims to be trustworthy and delivers answers that help as much as possible while steering clear of supporting harmful activities.
“We created evaluations to detect instances of deception and trained the model to be honest,” Beutel explained.
Beutel said ChatGPT-5 produces “safe completions,” giving high-level information that people can’t misuse to cause harm.
This week, the company also unveiled two new AI models—available for free download and open to modification by users—in a bid to compete with similar products from rivals.
The launch of these “open-weight” models comes amid pressure on OpenAI to share more about its technology, reflecting its nonprofit roots.
Read the original article on: Sciencealert
Read more:Enabling Data Storage to Evolve Alongside the AI Revolution