用 Jev 构建代码库分类器
开发者分享用 Jev 构建代码库分类器,认为这可能解决智能体生成过度工程化代码的问题,并询问下一步测试方向。
帖子对比了 Claude 与 Jev:Claude 需要反复读取全部内容来验证,循环越多成本越高;Jev 每百万输入令牌仅需 $0.042,输出免费,直接返回决策,而不是生成更多文本。
YOUR VERIFIER STILL COSTS LIKE A FULL MODEL Claude writes. then Claude reads the whole pile again to say yes or no. every extra loop turn makes that pile bigger. Jev: $0.042 per 1M input tokens. output tokens free. it returns a decision, not another essay. same job. wrong tool