搭终端演示 Grok 4.7 写作 + Jev 做路由决策的成本拆分
作者对比 Grok 4.7 与 GPT-5.6 Sol Max 的跑分和价格，并称每次路由、重试判断和「要不要调 LLM」都由 Jev 以 0.0003 美元一次处理，做了实时显示成本分配的终端 demo。
Grok 4.7 matches GPT-5.6 Sol Max on real benchmarks and costs half as much to run DeepSWE v1.1: 71.0% vs 72.7%. basically the same model input tokens: $2 vs $4 - half the price vs Fable 5.1 Max: $2 vs $10 - a fifth of the price and that's before Jev touches anything every routing call, every retry check, every "should this even go to the LLM" decision - Jev handles for $0.0003 a call, so the $2/$6 only fires when something actually needs generation built a terminal that runs this stack live - Grok 4.7 for the writing, Jev for everything else, watch the cost split happen in real time full build below ↓
这条是引用着另一条写的：x.com/solst1ne/status/2101724066369654935