Jev 在 System One 工作流评测上快 193.6 倍

Fluixo @fluixoo原推发布于 2026.09.22

在 X 上打开原推

193.6× FASTER. That is TypeSafe’s reported result on its own System One workflow eval not a universal model ranking. The interesting part is why the gap exists. LLMs generate strings token by token. Jev gives up free-form text and returns bounded decisions in parallel. Less freedom. Less bullshit. A radically different cost curve. If the independent benchmarks hold, agent architecture is about to change.

这条是引用着另一条写的：x.com/fluixoo/status/2101672915984888306