openai

OpenAI is rolling GPT-6 and Intelligent UI out across every ChatGPT tier, reaching 1.2 billion weekly users. The quiet shift is not the model — it is that the AI now decides whether your answer is text, a chart, a form or a working tool, and why that changes the deal.

OpenAI's new financial-services product ships with licensed data built in. But its three benchmarks improved by 3, 9.7 and 34 points — and the lopsidedness reveals what the product is really selling.

A question mathematicians left open for about 90 years — can smooth fluid motion break down on its own? — was answered in 88 hours by roughly 10,000 AI agents. Here is the problem in plain English, what OpenAI actually proved, and why nobody is claiming the $1 million.

OpenAI shipped GPT-6 Astra with near-perfect scores on three of its four flagship benchmarks. The scores are the least interesting thing about it — the price went up 2.5x, computer use is still the frontier, and OpenAI rates the model Critical for cyber capability.

OpenAI’s new ChatGPT Work agent takes a goal, works across your apps and files for hours, and hands back finished work. The most revealing detail is what it’s built on — a coding engine — and what that says about where AI is heading.

OpenAI’s GPT-Live models turn ChatGPT’s voice mode from a walkie-talkie into a phone call: the AI listens and speaks at the same time, hands hard questions to a frontier model behind the scenes, and is rolling out to everyone. Here’s what actually changed.