传闻 OpenAI 内部 AGI 模型 Bel 训练了 GPT-6 Astra
爆料称 OpenAI 有个叫 Bel 的内部模型，参数超 10 万亿、据称解出纳维-斯托克斯难题，还用来训练了 GPT-6 Astra；OpenAI 已确认存在一个比 Astra 更强的内部模型。
🚨 Wait… Bel might be OpenAI’s next fucking weapon. 👀 OpenAI reportedly has an internal AGI model called Bel. And the claims around this thing are fucking wild: → 10T+ parameters → Designed around AGI-level capabilities → Reportedly solved the Navier–Stokes Millennium Prize problem → Allegedly used to train GPT-6 Astra → Estimated 80+ hour reasoning horizon → Faster and Cheaper → Could reportedly arrive on devday or before 2027 And now OpenAI has confirmed something VERY important: An internal model significantly more capable than GPT-6 Astra solved the Navier–Stokes problem. OpenAI hasn't called that model Bel. But suddenly, the Bel leak doesn't sound nearly as fucking crazy. If even half of these Bel details are real, OpenAI may be sitting on something WAY beyond Astra. 👀