Speculation: OpenAI shelved Astra to make room for a stronger Bel-class model
One reading holds that rather than rush Astra out to beat Opus, OpenAI would rather spend that compute on a stronger internal model while showing regulators a safety-first stance.
What I think happened is that OpenAI basically “cleared the distractions out of the way.” Spending too much time on GPT-6.1 Astra right now may simply not be worth it if that same effort can go into the much more powerful Bel-class systems instead. From an engineering perspective, it makes more sense to focus on the thing that could genuinely shake Anthropic’s entire lineup rather than ship a faster model just to beat Opus. Because let’s be real, Fable 5.5 is sitting there waiting to bury OpenAI if they don’t respond hard enough by December. And strategically, the sequence makes sense: tell regulators you’re taking safety extremely seriously, spend that time shaping something much more powerful internally, then ship it once it’s actually safe enough. Everyone wins. And this lines up with something else I’ve heard for a while now: OpenAI has an internal model that makes the current Astra look like a weak little silicon baby by comparison.