GPT-6 Astra

GPT-6 Astra is OpenAI's newest AI model. What sets it apart is that it doesn't just handle text and code — it also judges things you can only really tell by seeing and interacting with them, like how objects are arranged in a game space, how the screen reacts, and what controls feel like.

Think of it like an architect who, given just one bare, textureless blueprint, can build several finished buildings with completely different moods from that same blueprint at once. In fact, the game startup PlayCo gave the model a single unfinished skeleton prototype and had it produce three themed prototypes from it at the same time — and most came out usable on the first try.

That said, this model is reportedly surrounded by safety controversy even before release. Its safety rating — which measures how well it handles risky requests — has raised concerns, so the actual features and restrictions available can vary depending on who is accessing it and how.

In articles, it tends to show up attached to a specific model name to describe performance gains, as in "switching to GPT-6 Astra cut manual work by 50% compared to before." However, the article also notes that such results came from a special access setup OpenAI granted to a defense partner, so the performance and limitations may differ from the default GPT-6 Astra that regular developers access via the API.