传闻：OpenAI 内部模型 Bel 强于 Astra
爆料称 OpenAI 内部系统 Bel 解出纳维-斯托克斯千禧年难题，能力超过 GPT-6 Astra，参数或超 10 万亿。
🚨 Bel Leaks: OpenAI's Model That Outpaces Astra >OpenAI confirmed an internal system "notably more capable than GPT-6 Astra" cracked the Navier-Stokes Millennium Prize Problem >One of just seven unsolved millennium-level math problems, open for roughly nine decades >OpenAI isn't going after the $1M prize money, treating it more as proof of how fast capability is scaling >Not everyone's convinced — an Anthropic researcher reportedly had a similar solution in progress independently, sparking a credit dispute >The internal codename floating around: Bel >Could end up as the base layer for an AGI-threshold system >Rumored to run past 10T total parameters Could this land sometime before 2027, once GPT-6.1 Astra clears the runway?