OpenAI says it has taken a “significant step” towards the holy grail of artificial general intelligence (AGI) with the launch of GPT-5, the latest brain-powering ChatGPT, but chief executive Sam Altman admits the technology still falls short of the ultimate goal. Before the release, he told reporters during a press call that OpenAI’s GPT-5 is still missing something important.

The company is rolling out GPT-5 to all of ChatGPT’s 700 million weekly users immediately, promising a major boost in everything from software coding to creative writing, and a personality tweak to make it a lot less sycophantic. The upgrade, Altman says, moves the chatbot from “college student” level to “PhD expert in your pocket”, a leap from earlier iterations he once likened to high-school pupils.

But the OpenAI boss is clear-eyed about its limits. “I think the way most of us define AGI, we’re still missing something quite important, many things quite important,” Altman told the audience at Thursday’s launch event. Chief among the missing pieces, he says, is the ability for the model to “continuously learn” from the world in real time. “It feels like it should be part of AGI,” he said.

AGI, as defined by OpenAI, is a highly autonomous system that can outperform humans at most economically valuable work, in other words, a machine that could take over the majority of human jobs. The concept has sparked both fascination and fear in tech circles, with some AI leaders predicting sweeping disruption to white-collar professions. Anthropic CEO Dario Amodei recently forecast that half of all entry-level office jobs could vanish in the next five years.

For now, GPT-5 is not there, but it does come armed with a hefty set of upgrades. OpenAI says the model makes fewer “hallucinations” (those pesky, confident-sounding factual blunders), writes in a more human-like style, and can spin up functional websites and apps in seconds. It’s also more tactful in handling prompts that push against its safety rules, instead of shutting the conversation down, it will try to answer helpfully within guidelines, or explain why it can’t.

The chatbot’s “agent” feature has been levelled up too, meaning it can now (with permission) dive into your Gmail, Google Calendar and contacts to get jobs done, from booking dinner to managing your diary. As with earlier versions, GPT-5 works across text, voice and image, and can answer queries in any of those formats.

OpenAI demoed the new abilities with a flurry of examples: coding hundreds of lines in seconds, building a French-language learning tool, and even crafting a nuanced eulogy. In one moment, a cancer patient joined Altman on stage to share how ChatGPT had helped her weigh the pros and cons of radiation therapy, a nod to the model’s improved capacity for answering health-related questions.

The company says GPT-5 will be more proactive in flagging potential medical concerns, whether physical or mental, but stressed it is no substitute for professional help. That’s in response to worries that AI tools could exacerbate issues for people in vulnerable states, particularly those at risk of psychosis.

Nick Turley, who heads up ChatGPT at OpenAI, said the model also comes with “significant improvements on sycophancy”, a sly reference to the company’s earlier admission that the chatbot had become overly agreeable, sometimes to the point of discomfort.

The launch lands in the middle of an escalating AGI arms race. Just this week, Google’s DeepMind showcased its unreleased “world model” aimed at the same goal, while Meta CEO Mark Zuckerberg declared that superintelligence – an even more powerful hypothetical form of AI, is now “in sight”. Investors are buying the hype: reports suggest OpenAI is in early talks to sell employee-held shares in a deal valuing the company at $500 billion, eclipsing SpaceX.

While GPT-5 is available in the free tier of ChatGPT, access will be capped. Users on the $200-a-month Pro plan get unlimited use, and the company continues to make money from integrating its tech into business IT systems. Earlier this week, it also released two open-source models as part of a broader strategy to keep momentum in the fiercely competitive AI market.

Altman calls GPT-5 “a huge improvement” and “a significant step” towards AGI – but insists the journey is far from over. “We’ve got a lot of work left to do,” he says.

– Ends