The ability to utilize preclinical models to predict the clinical toxicity of chimeric antigen receptor (CAR) T cells in solid tumors is tenuous, thereby necessitating the development and evaluation of gated systems. Here we found that murine GD2 CAR-T cells, specific for the tumor-associated antigen GD2, induce fatal neurotoxicity in a costimulatory domain-dependent manner. Meanwhile, human B7H3 CAR-T cells exhibit efficacy in preclinical models of neuroblastoma. Seeking a better CAR, we generated a SynNotch gated CAR-T, GD2-B7H3, recognizing GD2 as the gate and B7H3 as the target. GD2-B7H3 CAR-T cells control the growth of neuroblastoma in vitro and in metastatic xenograft mouse models, with high specificity and efficacy. These improvements come partly from the better metabolic fitness of GD2-B7H3 CAR-T cells, as evidenced by their naïve T-like post-cytotoxicity oxidative metabolism and lower exhaustion profile.