Can General-purpose Multimodal AI Reliably Measure Cobb Angles? A Head-to-head Diagnostic Agreement Study of GPT-5.1 and Gemini 3.1 Pro in Adolescent Idiopathic Scoliosis.

General-purpose multimodal artificial intelligence (AI) models can interpret radiographs through natural-language prompts, but their quantitative agreement and anatomical traceability in adolescent idiopathic scoliosis remain uncertain. We compared GPT-5.1 and Gemini 3.1 Pro with a joint expert-consensus reference on 125 anonymized standing coronal whole-spine radiographs. Each model evaluated each image once in an independent chat session using the same standardized prompt; only the first response was retained. Outcomes included single-measure absolute-agreement intraclass correlation coefficients (ICCs), mean absolute error (MAE) relative to the reference, Bland-Altman analysis, error-threshold performance, categorical agreement, and end-vertebra concordance. Gemini 3.1 Pro showed higher ICC (0.842 versus 0.587), lower mean bias (3.46° versus 8.68°), lower MAE (5.06° versus 9.62°; paired P < 0.001), and more measurements within 5° of the reference (61.6% versus 24.8%). Agreement with reference categories was 80.0% versus 52.8% for angle-derived severity and 78.4% versus 52.8% for prompt-constrained management categories (both paired P < 0.001). Exact upper end-vertebra agreement was 14.4% versus 8.8%, and exact lower end-vertebra agreement was below 20% for both models. Gemini 3.1 Pro performed better under the tested single-session protocol, but neither model demonstrated sufficient case-level agreement and end-vertebra concordance to support autonomous Cobb-angle measurement. These results are relative to expert consensus and do not establish absolute measurement accuracy or model repeatability. Expert verification remains necessary.