ClinFusion researchers report gains across 2D and 3D medical benchmarks

Researchers behind the ClinFusion preprint report that their medical AI system outperformed leading open-source medical models on 20 out of 24 benchmarks. The system combines processing of 2D images and native 3D medical images in one encoder, supporting visual questions, report generation and instruction following. The researchers also report better multimodal results than GPT-5.2 and Gemini-3-Flash on 13 out of 16 benchmarks. In a blinded evaluation, board-certified radiologists ranked ClinFusion's reports highest. Those findings concern benchmark and report evaluations, rather than demonstrated clinical deployment.