Sam Altman Confirms OpenAI’s Astra Model Aiming for AGI Breakthrough by Year-End

OpenAI wants AGI by December. Sam Altman said so himself — and he’s not being subtle about it.

Altman made clear the company hasn’t actually hit the AGI threshold yet, but the plan is to have an internal system that could qualify as one before the year wraps up. The vehicle for that push is a model called Astra, built specifically to handle complex research tasks without someone hovering over it the whole time. Think of it less like a chatbot and more like a junior colleague who can set up experiments, write code, and keep working while you sleep. That’s the pitch, anyway.

Not everyone inside OpenAI is measuring the same distance to the finish line.

Mark Chen, OpenAI’s research director, put the company at roughly “80% of the way” to AGI — his words. Greg Brockman, one of OpenAI’s co-founders, went further, saying AGI could basically be seen as having already emerged during this period. Two very different reads on the same moment. Worth noting: neither number comes from a standardized scientific benchmark. These are internal assessments, shaped by the people doing the work and, presumably, rooting for it to succeed.

What Astra Actually Does

Jakub Pachocki, OpenAI’s chief scientist, said Astra can automate tasks that a beginner researcher would normally handle — experiment setup, code writing, that kind of thing. The idea is that a system like Astra could speed up the development of future AI models through what amounts to a recursive loop: AI helps build better AI. Faster cycles, bigger leaps, compounding returns on research time.

But here’s the catch. OpenAI hasn’t released detailed technical papers on Astra. No external evaluations have been published. So right now, the system’s capabilities are pretty much defined entirely by the people who built it, which is not exactly an independent audit. The company says it’s being cautious — safety protocols are being tightened, which could push back the release date. That’s a reasonable explanation. It’s also convenient.

OpenAI’s own charter defines AGI as a system that surpasses human abilities across economically valuable tasks. Astra agents are supposed to be able to work on complex problems without constant supervision, running extended operations that look more like a virtual colleague than a search tool. Whether Astra clears that bar is, at this point, an open question — one that OpenAI’s internal team answers one way and the broader scientific community can’t yet answer at all.

Related: Jeonbuk Bank Partners with Ripple to Challenge SWIFTs Cross-Border Payment System

Microsoft’s Veto and the Money on the Table

There’s a harder constraint sitting in the background, and it has Microsoft’s name on it.

OpenAI’s partnership with Microsoft includes a clause that any official AGI declaration has to go through an independent expert panel review. Altman’s personal conviction won’t cut it. The panel has to sign off. Until that happens, the financial arrangements stay as they are — profit sharing between the two companies, plus Microsoft’s rights over OpenAI’s models, all of which run through 2032 and hinge on that AGI confirmation.

So the stakes aren’t just scientific. They’re financial, structural, and tied to a legal agreement between two of the most powerful companies in tech. An AGI call that jumps the gun could blow up the whole arrangement. That’s probably why OpenAI is moving carefully, even as Altman talks openly about year-end timelines.

The company says human and AI collaboration will continue until at least 2028 regardless — which is a pretty big qualifier sitting right next to all the AGI talk. It’s not a contradiction exactly, but it does complicate the picture. You can believe AGI is close and still plan for years of human oversight. Those two things can coexist. But it does make the “AGI by December” framing feel a bit more aspirational than operational.

What’s Missing From the Story

No public timeline for Astra’s release has been set. OpenAI hasn’t said when external evaluations might happen or who would conduct them. The expert panel that Microsoft’s partnership requires — unclear who’s on it, unclear how it convenes.

What’s clear is that OpenAI is moving fast internally and slowly publicly. Astra is being developed, safety reviews are happening, and the company believes it’s close. But “close” by internal metrics, without external validation, is a provisional claim. It might be right. It might not be. The absence of published technical data makes it genuinely hard to know.

Altman has staked his credibility on this timeline before. AGI predictions have a long history of sliding. The Microsoft panel requirement means the finish line isn’t just OpenAI’s to draw.

Profit-sharing arrangements between OpenAI and Microsoft stay in place until the expert panel confirms otherwise.