Background: Cognitive impairments exist in a large proportion of remitted patients with bipolar disorder (BD). However, no study has investigated the cognitive trajectories across neurocognitive subgroups of patients or their unaffected first-degree relatives (UR). Methods: Newly diagnosed BD patients, UR and healthy controls (HC) underwent comprehensive cognitive testing at baseline and at 16-months follow-up. Hierarchical cluster analysis was conducted to identify homogeneous subgroups of patients based on their neurocognitive profile at baseline. Cognitive change across subgroups of patients and UR was assessed with linear mixed-model analyses. Results: Data from baseline and follow-up were collected from 152 patients, 53 UR and 135 HC. Patients were clustered into three discrete neurocognitive subgroups: ‘cognitively normal' (43%), ‘mild-moderately impaired’ (33%) and ‘globally impaired’ (24%). While ‘mild-moderately impaired’ patients and HC showed normative cognitive improvement over time in global cognition (p < .001), ‘globally impaired’ patients showed greater improvement than all groups (p < .001), whereas ‘cognitively normal' patients showed a lack of normative improvement (p = .17). UR of impaired patients showed a lack of normative improvement in executive functions (p = .01). ‘Globally impaired’ patients also presented with stable impairments in facial expression recognition and emotion regulation. Limitations: Follow-up data was available for 62% of participants, possibly reflecting a selection bias. Conclusions: The greater cognitive improvement in ‘globally impaired’ patients partly speaks against neuroprogression. However, the lack of normative improvement in ‘cognitively normal' patients could indicate negative effects of illness. Further follow-up assessments are warranted to clarify whether lack of normative improvement in executive function in UR represents an illness risk-marker.