WHAT NEW IN X10.31 SAT

X 上的 AI 合集档案 · 22 个合集 · 62,909 条

gpt-6POST

观点:Anthropic 就算跑分赢 Astra 也难赢

作者认为 Anthropic 连现有模型满负荷都撑不住,OpenAI 的算力和可用性优势比跑分更关键。

Ananth
@Ananth7e2026.09.14收录原推发布于 2026.09.09
Ananth @Ananth7e原推发布于 2026.09.09
在 X 上打开原推

Anthropic could beat GPT-6 Astra on every benchmark and still lose even if anthropic launches an "astra killer" next, whether that's fable 5.2 or something else, i don't think the model quality alone fixes their problem. they can barely serve fable 5.1 at full usage today. so how are they supposed to serve a fresh pretrain, potentially a bigger model to everyone at scale? that's where openai has a huge advantage. astra is already an absolute beast, and 100% of your usage actually counts toward the model you want to use. anthropic can ship something better on paper, but if users constantly hit caps, it doesn't matter nearly as much. unless anthropic magically finds a mountain of compute under the bed, the usability problem is still there.

x.com/ananth7e/status/2097656093094949184采集于 2026.09.14 15:53