gpt-6IMAGE
Astra 与 Opus 5.5 的 token 消耗对比
该帖称 Astra High 每任务约 1.2 万 token,Opus 5.5 需约 6.6 万,效率差 5.5 倍,并猜测 OpenAI 内部还藏着更强的教师模型。
VraserX e/acc
@VraserX29 天前收录原推发布于 2026.10.05

VraserX e/acc @VraserX原推发布于 2026.10.05
Token for token, GPT-6 Astra doesn’t just beat Opus 5.5. It annihilates it. Astra High uses ~12k tokens per task. Opus 5.5 needs ~66k. That’s 5.5× the token efficiency while Astra is still more intelligent pound for pound. Now think about what that implies. If Astra is the distilled model, what the hell was the teacher? OpenAI could have an internal mode lurking behind the curtain that makes Opus 5.5 and Fable 5.5 look obsolete before we’ve even seen it.
x.com/vraserx/status/2107094957035307174采集于 2026.10.06 10:03