OpenAI’s leadership is making a bold claim: the company is roughly 80% of the way to artificial general intelligence, and CEO Sam Altman expects an internal system he’d call AGI by the end of 2026. That’s according to a sprawling TIME report by Alex Heath, who spent over two weeks interviewing more than 20 executives, employees, investors, and rivals. Chief Research Officer Mark Chen put the number at 80%, while co-founder Greg Brockman framed this period as the moment AGI emerges. Altman himself was more careful: “not quite yet,” but close.
The centerpiece of this confidence is Astra, OpenAI’s upcoming model family first announced as an automated AI research intern. Chief Scientist Jakub Pachocki says Astra already hits that benchmark. Give it an experimental idea and it writes code in OpenAI’s codebase, runs the experiment, and reports back. Hand it a research paper and it handles work that would take a human researcher about a week. Altman’s pitch to customers goes further: “I expect this will be the first model where the model actually invents new things in a way that matters. That’s a very AGI-like thing.”
The underlying bet is recursive self-improvement — AI that helps build better AI. Some researchers see early signs that’s already happening. Others think it’s still far off. The deeper problem is definitional. OpenAI’s own AGI definition is heavily economics-driven: “highly autonomous systems that outperform humans at most economically valuable work.” Critics argue that’s a convenient benchmark for a company that needs to justify its valuation, and that genuine AGI requires robust world understanding that language models alone may never deliver.
Meanwhile, OpenAI faces a more mundane challenge: paying for all this. The company is merging Codex with ChatGPT in a project called “The Merge,” pushing ChatGPT Work toward “persistence and always-on execution,” according to product lead Thibault Sottiaux. Ads inside ChatGPT are scaling up after early tests, with CFO Sarah Friar noting that ad revenue could subsidize the 92% of consumer users who don’t pay. OpenAI is also testing Sponsored Agents, where clicking an ad drops users into a brand-hosted AI experience. And yes, hardware is coming: a puck-shaped device that senses its surroundings and talks via voice mode, expected early 2027, plus humanoid robots that Altman says OpenAI will “definitely” build.
