There is a paucity of data on the natural trajectory of outcomes in survivors of COVID-19 beyond 2 years after symptom onset, and no evidence exists on the effect of re-infection in people with long COVID symptoms. We aimed to investigate the 3-year health outcomes of COVID-19 survivors and the effect of omicron re-infection. In this single-centre, longitudinal cohort study, we recruited participants with confirmed COVID-19 who were discharged from the Jin Yin-tan hospital in Wuhan, China, between Jan 7 and May 29, 2020. Participants completed three follow-up visits at 6 months (June 16 to Sept 13, 2020), 1 year (Dec 16, 2020, to Feb 7, 2021), and 2 years (Nov 16, 2021, to Jan 10, 2022) since symptom onset (reported previously). At 1-year follow-up, community controls without a history of SARS-CoV-2 infection were recruited from two communities in Wuhan and at 2 years were matched (1:1) with survivors of COVID-19 who underwent pulmonary function tests. We did a 3-year follow-up from Feb 23, 2023, to April 20, 2023, after the omicron (B.1.1.529) wave in winter, 2022. All eligible survivors of COVID-19 and community controls matched at 2-year follow-up were invited to the outpatient clinic at the hospital to complete several face-to-face questionnaires, a 6-min walking test (6MWT), and laboratory tests. A subgroup of survivors of COVID-19 identified by stratified sampling on the basis of disease severity scale score during hospitalisation and community controls underwent pulmonary function tests. Survivors of COVID-19 who received high-resolution CT and showed abnormal lung images at 2-year follow-up were invited for another assessment. We identified participants with and without long COVID at 2 years. The primary outcomes were sequelae symptoms, omicron infection, lung function, and chest imaging at the 3-year follow-up. Of 1359 COVID-19 survivors who completed 2-year and 3-year follow-up, 728 (54%) had at least one sequelae symptom at 3 years after symptom onset and before omicron infection, mainly mild to moderate severity. During the omicron wave, participants with long COVID at 2 years had a significantly higher proportion of re-infection (573 [76%] of 753 vs 409 [67%] of 606 without long COVID; p=0·0004), pneumonia (27 [5%] of 568 vs seven [2%] of 403; p=0·012). 3 months after omicron infection, 126 (62%) of 204 survivors with long COVID at 2 years had newly occurring or worse symptoms, which was significantly higher than the proportion in the non-long COVID group (85 [41%] of 205; p<0·0001) and community controls (81 [40%] of 205; p<0·0001), and not significantly different between COVID-19 survivors without long COVID and matched community controls (85 [41%] of 205 vs 81 [39%] of 206; p=0·66). Re-infection was a risk factor for dyspnoea (odds ratio 1·36 [95% CI 1·04 to 1·77]; p=0·023), anxiety or depression (OR 1·65 [1·24 to 2·20]; p=0·0007), EuroQol visual analogue scale score (β -4·51 [-6·08 to -2·95]; p<0·0001), but not for reduced daily activity (0·72 [0·38 to 1·37]; p=0·32) at 3 years. Lung function of survivors at 3 years was similar to that of matched community controls. We found irregular line, traction bronchiectasis, subpleural lines and ground glass opacity at 3 years, but the volume ratio of lung lesion to total lung was only 0·2-0·3%. Most long COVID symptoms at 3 years were mild to moderate, with lung function recovering to levels of matched controls. Survivors with long COVID had a higher proportion of participants with re-infection and newly occurring or worse symptoms 3 months after omicron infection than those without long COVID. Re-infection had increased symptom occurrence but not increased reduced daily activity. Although the organ function of survivors of COVID-19 recovered over time, those with severe long COVID symptoms, abnormal organ function, or limited mobility require urgent attention in future clinical practice and research. Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences, National Natural Science Foundation of China.