Leak: OpenAI is distilling cheaper new variants from Astra
A claim that OpenAI is using GPT-6 Astra to produce new variants along the faster, cheaper path of 6.1 Sol, with a model codenamed Bel possibly not appearing until around Christmas.
OpenAI is still using GPT-6 Astra to help produce new variants, just like it did with GPT-6.1 Sol, which is nearly as capable as Astra, took about half the time to build, is roughly 5x more efficient, and dramatically cheaper. New variants are being developed right now, tuned and prepared for releases over the next few weeks, including some really impressive fine-tuning work for Chat on top of the models that already exist. Don’t get it twisted, “Bel” isn’t going to be released for no reason. The current expectation is that we may start seeing signs of it by Christmas, not in October. Until then, OpenAI may keep giving Anthropic the edge in raw performance while attacking on price, speed, and efficiency instead. Basically the same playbook they used against Opus 5.5.
OpenAI isn’t going to release Bel just because of Fable 5.5, trust me. Bel-class models are so powerful that it wouldn’t even make sense to use them to counter “just Fable 5.5.” And I say that with confidence, because OpenAI’s internal models are around 2-3x better than GPT-6 Astra. They’re deliberately holding those releases back for the right moment, and from what I’ve been able to gather, a big part of that has to do with “safety risks” at a level they haven’t dealt with before.