ChatGPT-5 debuts free for all users

OpenAI has unveiled a highly anticipated new version of its flagship ChatGPT, boasting major advancements in artificial intelligence as the global race to dominate the technology intensifies.

The latest model, ChatGPT-5, is being made available free to all users of the platform, which serves nearly 700 million people weekly, the company confirmed during a media briefing.

Co-founder and CEO Sam Altman described the upgrade as “clearly a model that is generally intelligent,” but acknowledged that it still falls short of achieving artificial general intelligence (AGI) — a system that can think like humans.

“This is not a model that learns continuously from new information as it is used, which is something I believe should be part of AGI,” Altman explained. “But its capabilities represent a huge leap forward.”

Industry experts have welcomed the release as a milestone in the AI era, predicting that increasingly powerful systems will transform work and daily life. Meta CEO Mark Zuckerberg, in a recent memo, wrote that the rapid pace of progress meant “superintelligence is coming into sight” and described this as the beginning of “a new era for humanity.”

Altman said OpenAI expects “orders of magnitude more gains” on the road to AGI, noting that achieving it requires “eye-watering” investment in computing power — something the company intends to continue.

Since ChatGPT’s initial launch in late 2022, tech giants including Amazon, Google, Meta, Microsoft, and Elon Musk’s xAI have invested billions into AI. Earlier this year, Chinese startup DeepSeek disrupted the sector with a high-performing model that uses less expensive chips.

Highlighting ChatGPT-5’s performance, Altman compared its evolution to human learning. “GPT-3 was like talking to a high school student — sometimes you got the right answer, sometimes not. GPT-4 felt like a college student. GPT-5 is the first time it feels like speaking to a PhD-level expert on any subject.”

He predicted that one defining feature of this new era would be “vibe-coding,” where the AI can build software on demand based on a user’s description.

British AI expert Simon Willison, who tested ChatGPT-5 early, described it as consistently competent. “It rarely messes up and often impresses me,” he said in a blog post, though he noted it didn’t feel like a dramatic leap from other leading models. Musk, however, claimed on X that his own Grok 4 Heavy AI was “smarter” than ChatGPT-5.

OpenAI’s safety research lead, Alex Beutel, said the model was trained to be truthful and avoid enabling harmful activities. “We built evaluations to measure deception and trained the model to be honest,” he said, adding that ChatGPT-5 focuses on “safe completions” that avoid giving dangerous, actionable details.

Alongside ChatGPT-5, OpenAI also introduced two open-weight AI models that users can download and modify for free, in a move seen as a challenge to rival offerings. The release comes as the company faces mounting pressure to reveal more about its technology, in line with its roots as a nonprofit.