Jev 决策层实测：193 倍快、444 倍便宜
称 Jev 读状态返回带校准概率的类型化答案，70–500ms 出结果，输入 $0.042/M、输出免费，1 万次决策 $0.42，且 schema 保证不会返回代码没处理的分支。
this paper is f*cking brilliant Now Jev Engineering adds the decision layer every agent stack is missing LLM writes → Jev decides → code acts Jev is a System One model by TypeSafe. it doesn't generate text. it reads a state, answers questions you defined in advance, and returns typed values with calibrated probabilities three shapes of answer: choice: one winner from a closed list score: a position on your rubric noul: the probability that a statement holds the result: a typed answer in 70–500 ms, against 3–329 s for a frontier call. 193.6× faster and 444.6× cheaper end to end the crazy part is the price. $0.042 per million input tokens, and output tokens are free 10,000 decisions for $0.42 1,018 papers sorted for $0.08 it can't return a branch your code doesn't handle. zero type errors, by schema read the full paper + repo below bookmark it for future reference
这条是引用着另一条写的：x.com/noisyb0y1/status/2102309802072272956