Cognitive deficits have an important role in the neurodevelopment of schizophrenia and other psychotic disorders. However, there is a continuing debate as to whether cognitive impairments in the psychosis prodrome are stable predictors of eventual psychosis or undergo a decline due to the onset of psychosis. In the present study, to determine how cognition changes as illness emerges, we examined baseline neurocognitive performance in a large sample of helping-seeking youth ranging in clinical state from low-risk for psychosis through individuals at clinical high-risk (CHR) for illness to early first-episode patients (EFEP). At baseline, the MATRICS Cognitive Consensus battery was administered to 322 individuals (205 CHRs, 28 EFEPs, and 89 help-seeking controls, HSC) that were part of the larger Early Detection, Intervention and Prevention of Psychosis Program study. CHR individuals were further divided into those who did (CHR-T; n = 12, 6.8%) and did not (CHR-NT, n = 163) convert to psychosis over follow-up (Mean = 99.20 weeks, SD = 21.54). ANCOVAs revealed that there were significant overall group differences (CHR, EFEP, HSC) in processing speed, verbal learning, and overall neurocognition, relative to healthy controls (CNTL). In addition, the CHR-NTs performed similarly to the HSC group, with mild to moderate cognitive deficits relative to the CTRL group. The CHR-Ts mirrored the EFEP group, with large deficits in processing speed, working memory, attention/vigilance, and verbal learning (>1 SD below CNTLs). Interestingly, only verbal learning impairments predicted transition to psychosis, when adjusting for age, education, symptoms, antipsychotic medication, and neurocognitive performance in the other domains. Our findings suggest that large neurocognitive deficits are present prior to illness onset and represent vulnerability markers for psychosis. The results of this study further reinforce that verbal learning should be specifically targeted for preventive intervention for psychosis.