OpenAI unveiled GPT-5 on Friday, its most capable artificial intelligence model to date, featuring native multimodal reasoning, autonomous agent capabilities, and what the company describes as “near-human” performance on professional-level reasoning benchmarks.
The new model can browse the live web, write and execute code in a sandboxed environment, analyze images and video, and hold coherent multi-turn conversations that span hours without losing context. OpenAI CEO Sam Altman called it “a step change in what AI can do for people.”
“GPT-5 isn’t just better at answering questions — it can actually do things,” Altman said during a livestreamed demo. “It can plan a vacation, book flights, debug your code, and explain its reasoning every step of the way.”
Early benchmarks shared by OpenAI show GPT-5 scoring 94% on the bar exam, 92% on the USMLE medical licensing test, and 89% on a new “general reasoning” benchmark designed to test multi-step problem solving. The model also demonstrated significant improvements in reducing hallucinations, a persistent challenge for large language models.
The announcement sent ripples through the tech industry. Google and Anthropic are expected to respond with their own model updates in the coming weeks. Microsoft, OpenAI’s largest partner, said it would integrate GPT-5 into Copilot “within days.”
Privacy advocates raised concerns about the agent capabilities, noting that a model with web access and execution privileges could pose security risks if misused. OpenAI said it had implemented “extensive safety measures” and would roll out agent features gradually.
GPT-5 is available today for ChatGPT Plus subscribers, with API access priced at $15 per million input tokens.
Photo by Andrew Neel on Unsplash
As an Amazon Associate, we earn from qualifying purchases at no extra cost to you.