Why Polymarket Still Prices OpenAI AGI at 19% After Astra Drives

A robot just drove a car around a real cone course, and the prediction market barely moved. Polymarket still gives OpenAI a 19% chance of announcing AGI before 2027, even after that result landed.

The odds did jump 4 points once the DrivingBench numbers came out. Then $272,000 in trading volume settled in and said the crowd is not calling this the arrival of AGI — and the gap tells you something useful about how to read AI news.

What Happened

Astra drove a car in the real world and passed a cone course. Prediction markets treat that as a task milestone, not as AGI, and the betting volume shows a consensus most researchers seem to share.

The result proves something specific: mature vision-action multimodal capability. The car is not reading text, it is seeing a course and acting on it.

But the working definition of AGI asks for something much larger — autonomy across the majority of human work that carries real economic value. Clearing a cone course is nowhere near that bar, and the market prices the difference at 81%.

So the two things are not in conflict. Astra is not AGI. It is concrete evidence that modern LLMs are starting to touch the physics of the real world instead of staying inside text tokens.

Why It Matters

If you follow AI releases, the habit worth building is separating a demo from a definition. A robot that drives well is a real achievement, and a milestone for spatial agentic loops, but the research community reads it as one step rather than the finish line.

The market gives you a number to check that reading against. When the next dramatic demo shows up, look at where the odds actually sit before you decide what it means.