everyone's testing GPT-Astra on random 3D environments and sloppy games... so here's real feedback from actually working with it: - the most pleasant OpenAI model i've ever talked to, gives me Opus 4.5 vibes - exceptional at setting its own goals and prompting subagents through long horizon work - still behind Fable on frontend design - stupidly good at using the computer, this will open up insane workflows - much, much better at writing, looks like they finally cracked it - super efficient,…