Jev 在困难案例上落后于前沿 LLM

4/ The catch: they lose on hard cases. In one benchmark Jev scored 72.5% vs 84% for a frontier LLM. An independent test found decision models tie LLMs on safety classification and tool routing, and fall behind on open-ended tasks.