OpenAI confirms work on GPT-5, eyes superintelligence

If you thought GPT-4 was the ceiling for OpenAI, think again.

OpenAI CEO Sam Altman confirmed in an interview with the Financial Times that the company has officially started building its next-generation flagship model, GPT-5. Altman hinted that the model could potentially reach superintelligence, crossing the threshold into true artificial general intelligence.

The hunt for billions and private data

Building something beyond human-level reasoning doesn't come cheap, and OpenAI is already passing the hat.

Altman noted that the partnership with Microsoft is working smoothly, but made it clear that OpenAI will need a lot more cash from the Windows maker and other backers. Microsoft has already poured over $10 billion into the startup, yet Altman stressed that there is still a massive amount of compute to build out before reaching AGI.

To feed that compute engine, OpenAI is looking far beyond the standard web crawl.

Altman explained that training GPT-5 will require unprecedented volumes of data. That means blending publicly available online material with datasets purchased directly from companies, including private archives of long-form writing and written conversations that have never been widely indexed on the web.

As for when anyone will actually get to use it, do not hold your breath. Altman declined to offer a timeline or detail specific capabilities, admitting that nobody really knows what the model can do until training wraps up. Between raw training runs and extensive safety testing, the finish line remains months, if not years, away.

Why it matters

We think this announcement is as much about fundraising leverage as it is about frontier science.

For everyday users and developers, there is zero reason to pause current plans. GPT-4 remains the practical benchmark, and building robust workflows on existing tools is far more valuable right now than waiting on an unreleased supermodel. The operational cost and access friction for GPT-5, whenever it arrives, will almost certainly be steep.

For the broader industry, Altman is deliberately raising the ante. By signaling that the next leap requires billions in fresh capital and proprietary licensed data, OpenAI is squeezing smaller rivals out of the frontier tier. The frontier race is increasingly turning into an exclusive club where only hyperscalers can afford the table stakes.

There is an obvious downside to this posturing, though. Promising superintelligence keeps OpenAI in the headlines and loosens investor checkbooks, but it also paints a giant target on the company for regulators wary of existential risk. If the final model delivers solid reasoning gains rather than a genuine sci-fi breakthrough, the gap between Altman's hype and reality could prove costly.