To assess long-term efficacy and safety of open-source automated insulin delivery (AID) in children and adults (7-70 years) with type 1 diabetes. Both arms of a 24-week randomized controlled trial comparing open-source AID (OpenAPS algorithm within a modified version of AndroidAPS, preproduction DANA-i™ insulin pump, Dexcom G6 continuous glucose monitor) with sensor-augmented pump therapy (SAPT), entered a 24-week continuation phase where the SAPT arm (termed SAPT-AID) crossed over to join the open-source AID arm (termed AID-AID). Most participants (69/94) used a preproduction YpsoPump insulin pump during the continuation phase. Analyses incorporated all 52 weeks of data, and combined between-group and within-subject differences to calculate an overall "treatment effect" of AID versus SAPT. Mean time in range (TIR; 3.9-10 mmol/L 70-180 mg/dL) was 12.2% higher with AID than SAPT (95% confidence interval CI 10.4 to 14.1;  < 0.001). TIR was 56.9% (95% CI 54.2 to 59.6) with SAPT and 69.1% (95% CI 67.1 to 71.1) with AID. The treatment effect did not differ by age (  = 0.39) or insulin pump type (  = 0.37). HbA1c was 5.1 mmol/mol lower 0.5% with AID (95% CI -6.6 to -3.6;  < 0.001). There were no episodes of diabetic ketoacidosis or severe hypoglycemia with either treatment over the 48 weeks. Six participants (all in SAPT-AID) withdrew: three with hardware issues, two preferred SAPT, and one with infusion-site skin irritation. Further evaluation of the community derived automated insulin delivery (CREATE) trial to 48 weeks confirms that open-source AID is efficacious and safe with different insulin pumps, and demonstrates sustained glycemic improvements without additional safety concerns.