OpenAI's New Astra AI Can Build 3D Worlds and Drive Your Computer
Testers call it the biggest leap yet — and they're now running two AIs at once.
OpenAI has released a new AI model called GPT-6 Astra, and the people who got early access are unusually excited. Reviewers describe the jump from the previous version as much bigger than the last upgrade — a rare thing in a field where each new model usually feels like a small step. Astra's standout skill is what one reviewer calls "ambitious projects": 3D scenes, games, and anything visual or interactive.
Why should you care? Astra is also very strong at "computer use" — meaning it can look at your screen and click, type, and fill in forms for you, like a junior assistant sitting at your desk. It's good at "subagent coordination," which is one AI bossing around a team of smaller AIs to finish a big job. So the practical promise is real: tasks that took you an afternoon of tedious clicking could shrink to a few minutes of asking.
Here's the honest catch. Astra is not better at everything. On ordinary coding, it's an improvement but not a revolution. And the two big AI families still feel different — a rival model from Anthropic (nicknamed "Fable 5.1" in the review) remains the favorite for long, thoughtful back-and-forth discussions and editing. If you already prefer one company's style, that preference still holds.
There's also a bigger conversation. For the first time, serious people argued about whether a model like this counts as AGI — an AI as capable as a human at nearly every task. The reviewer says no, and warns against slapping that label on too early, but admits it's no longer a silly question. Meanwhile, OpenAI has already hinted it has an even stronger model waiting in the wings, and the rollout to regular users is happening gradually, not all at once.
- Astra's biggest wins are 3D, games and visual work — not plain coding, where it's good but not a revolution
- "Computer use" means it can operate software for you: clicking, typing and filling in forms on your screen
- Testers recommend running two AIs side by side and comparing answers, rather than trusting one
Why It Matters
Your most tedious computer chores — slides, forms, rough game prototypes — could soon be done for you in minutes.