Igor Pogany tests GPT-6 Astra on 20 examples selected to move beyond benchmark scores. The demonstrations span playable games, browser automation, research, web development, slide creation, document work and video production, with each result judged by whether it produces something usable rather than merely sounding convincing.
The strongest examples combine several modes of work. Astra researches unfamiliar subjects, operates websites, creates interactive artifacts and revises outputs after inspecting them. Pogany also highlights cases where long-horizon execution matters, because the model must keep track of intermediate decisions instead of solving a single isolated prompt.
Pogany's broader conclusion is that Astra is most valuable when a task has a clear outcome and enough room for the model to plan, use tools and check its own work. He also shows uneven quality across examples, so the video treats the model as a powerful operator that still benefits from deliberate task design and human review. Course and product promotions are omitted.
Watch on YouTube
