Background: When patients with type 2 diabetes start their first injectable therapy, clinicians can choose between glucagon-like peptide-1 (GLP-1) receptor agonists and basal insulins. In DURATION-3, exenatide once weekly was compared with insulin glargine (henceforth, glargine) as first injectable therapy. Here, we report the results of the final 3-year follow-up. Methods: DURATION-3 was an open-label randomised trial done between May 13, 2008, and Jan 30, 2012. Patients with type 2 diabetes aged 18 years or older were enrolled at 72 sites worldwide. They were eligible when they had suboptimum glycaemic control (HbA1c 7·1-11·0% [54-97 mmol/mol]) despite maximum tolerated doses of metformin alone or with a sulfonylurea for at least 3 months, a stable bodyweight for at least 3 months, and a BMI of 25-45 kg/m2 (23-45 kg/m2 in South Korea and Taiwan). Patients were randomly assigned (1:1) by computer-generated random sequence with an interactive voice-response system (block size four, stratified by country and concomitant therapy) to once-weekly exenatide (2 mg subcutaneous injection) or once-daily glargine (titrated to target) to be given in addition to their existing oral glucose-lowering regimens. The primary efficacy measure at 3 years was change in HbA1c from baseline in patients given at least one dose of the assigned drug (ie, analyses by modified intention to treat). Patients, investigators, and data analysts were not masked to treatment assignment. This trial is registered with ClinicalTrials.gov, number NCT00641056. Findings: 456 patients underwent randomisation and received at least one dose of the assigned drug (233 given exenatide, 223 glargine). At 3 years, least-squares mean HbA1c change was -1·01% (SE 0·07) in the exenatide group versus -0·81% (0·07) in the glargine group (least-squares mean difference -0·20%, SE 0·10, 95% CI -0·39 to -0·02; p=0·03). Transient gastrointestinal adverse events characteristic of GLP-1 receptor agonists were more frequent with exenatide than glargine (nausea: 36 [15%] of 233 patients vs five [2%] of 223; vomiting: 15 [6%] vs six [3%]; diarrhoea: 32 [14%] vs 15 [7%]), although frequency of these events did decrease after week 26 in the exenatide group. The proportion of patients who reported serious adverse events in the exenatide group (36 patients [15%]) was the same as that in the glargine group (33 [15%]). The exposure-adjusted rate of overall hypoglycaemia was three times higher in patients given glargine (0·9 events per patient per year) than in those given exenatide (0·3 events per patient per year). Interpretation: Efficacy of once-weekly exenatide is sustained for 3 years. GLP-1 receptor agonists could be a viable long-term injectable treatment option in patients with type 2 diabetes who have not yet started taking insulin. Funding: Amylin Pharmaceuticals and Eli Lilly.