The paper proposes a general method for constructing confidence intervals and statistical tests for single or low-dimensional components in high-dimensional models. The method is based on inverting the Karush-Kuhn-Tucker (KKT) conditions of the lasso estimator, which yields a nonparse estimator with a Gaussian limiting distribution. The authors analyze the asymptotic properties of this method and establish its asymptotic optimality in terms of semiparametric efficiency. The method is applicable to linear models and can be extended to generalized linear models with convex loss functions. The paper includes theoretical results for Gaussian, sub-Gaussian, and bounded correlated designs, and discusses the consistency of the lasso for nodewise regression as an inverse of the matrix of second-order derivatives. The authors also provide empirical results to support their theoretical findings.The paper proposes a general method for constructing confidence intervals and statistical tests for single or low-dimensional components in high-dimensional models. The method is based on inverting the Karush-Kuhn-Tucker (KKT) conditions of the lasso estimator, which yields a nonparse estimator with a Gaussian limiting distribution. The authors analyze the asymptotic properties of this method and establish its asymptotic optimality in terms of semiparametric efficiency. The method is applicable to linear models and can be extended to generalized linear models with convex loss functions. The paper includes theoretical results for Gaussian, sub-Gaussian, and bounded correlated designs, and discusses the consistency of the lasso for nodewise regression as an inverse of the matrix of second-order derivatives. The authors also provide empirical results to support their theoretical findings.