OpenAI GPT-6 Internal Demo Leaked Ahead of Thursday Launch
nashnova research
Developers posted demos of a model codenamed Astra on X, widely believed to be GPT-6; OpenAI has not confirmed, but the formal launch window points to this Thursday.
What do the leaked demos actually show?
A group of developers claiming access to an OpenAI internal checkpoint posted test videos of a model codenamed "Astra" on X.
The standout capability is 3D asset generation — producing three-dimensional objects from a single text prompt: a spaceship with engine sound effects, a castle with full interior structure, and a "Lego-style" pelican bicycle.
Community feedback says the model's reasoning intensity is notably above existing models, with detail control that barely requires human correction.
This means → if the demos are real, AI has jumped from "generate an image" to "generate a usable 3D scene" — an entirely different dimension.
Can these demos be trusted?
Every leaked demo is a self-reported internal-checkpoint result from community developers; OpenAI has not claimed any of them.
In plain terms = there is no official endorsement — credibility rests entirely on community reputation, so treat the information with caution.
However, X user @Lentils80 claims OpenAI appears to be expanding Astra's internal testing scope and has opened a new test codename, "mozaik-alpha-fdm," to select developers.
This reflects that even without verified demos, the spreading of test codenames suggests OpenAI is actively pushing something significant internally.
Thursday launch — what is the market watching?
The formal GPT-6 launch window is widely expected to be this Thursday.
The market's two core questions: ① Will OpenAI confirm Astra is GPT-6? ② How large is the gap between the shipping product and the leaked demos?
This means → Thursday is not just a product launch — it is an expectations calibration event. The demos raised the bar; the official release has to clear it.
市场有风险，内容仅供研究参考，不构成投资建议。