Objectives To evaluate the cost‐effectiveness of the Geriatric Care Model (GCM), an integrated care model for frail older adults based on the Chronic Care Model, with that of usual care. Design Economic evaluation alongside a 24‐month stepped‐wedge cluster‐randomized controlled trial. Setting Primary care (35 practices) in two regions in the Netherlands. Participants Community‐dwelling older adults who were frail according to their primary care physicians and the Program on Research for Integrating Services for the Maintenance of Autonomy case‐finding tool questionnaire (N = 1,147). Intervention The GCM consisted of the following components: a regularly scheduled in‐home comprehensive geriatric assessment by a practice nurse followed by a customized care plan, management and training of practice nurses by a geriatric expert team, and coordination of care through community network meetings and multidisciplinary team consultations of individuals with complex care needs. Measurements Outcomes were measured every 6 months and included costs from a societal perspective, health‐related quality of life (Medical Outcomes Study 12‐item Short‐Form Survey (SF‐12) physical (PCS) and mental component summary (MCS) scales), functional limitations (Katz activities of daily living and instrumental activities of daily living), and quality‐adjusted life years based on the EQ‐5D. Results Multilevel regression models adjusted for time and baseline confounders showed no significant differences in costs ($356, 95% confidence interval = −$488–1,134) and outcomes between intervention and usual care phases. Cost‐effectiveness acceptability curves showed that, for the SF‐12 PCS and MCS, the probability of the intervention being cost‐effective was 0.76 if decision‐makers are willing to pay $30,000 per point improvement on the SF‐12 scales (range 0–100). For all other outcomes the probability of the intervention being cost‐effective was low. Conclusion Because the GCM was not cost‐effective compared to usual care after 24 months of follow‐up, widespread implementation in its current form is not recommended.