Aims: To improve a real-time AI model(1) for both detection of laryngeal lesions and subsequently classify them as malignant or benign. To benchmark diagnostic performance against experienced clinicians in Head and Neck Oncology Centers (HNOCs). Methodology: For this multicenter diagnostic accuracy study, the AI model was trained on 597 endoscopic videos (collected 2012-2025) showing malignant, benign, or no glottic laryngeal lesions, with an internal test set of 356 videos. One in ten frames was extracted, manually annotated, and labeled using reference diagnoses. A prospective dataset included 98 consecutive patients referred to two HNOCs for suspected laryngeal cancer (2024-2025), processed identically to form a test set. For benchmarking, clinicians classified lesions at the video level as benign or malignant. Model performance in lesion detection was assessed using sensitivity and positive predicted value (PPV) against annotated bounding boxes. Frame-level predictions were aggregated to video-level outcomes and compared with clinician assessments. Sensitivity and specificity, and differences between AI and clinicians were analyzed (McNemar’s test). Results: Detection sensitivity and PPV were 0.78 (95%CI 0.74-0.82) and 0.86 (95%CI 0.84-0.88) on the internal test set and 0.81 (95%CI 0.76-0.85) and 0.91 (95%CI 0.89-0.93) respectively on the prospective test set. Comparing the AI model with HNOC clinicians, the AI model achieved a sensitivity of 0.97 (95%CI 0.92-1.00) and a specificity of 0.56 (95%CI 0.42-0.69), whereas for clinicians this was 0.89 (95%CI 0.80-0.97) (p = 0.10) and 0.65 (95%CI 0.52-0.79) (p = 0.17), respectively. Conclusion: This improved real-time AI model showed at least similar performance to clinicians in HNOCs, highlighting its potential for decision support among less experienced clinicians. 1. Wellenstein DJ, Woodburn J, et al. Detection of laryngeal carcinoma during endoscopy using artificial intelligence. Head Neck. 2023 Sep;45(9):2217-2226