United Imaging Intelligence Launches World's First Open-Source LLM for Medical Videos

United Imaging Intelligence Launches uAI NEXUS MedVLM

In an unprecedented move, United Imaging Intelligence (UII) has unveiled the uAI NEXUS MedVLM, distinguished as the first open-source large language model (LLM) specifically designed for medical videos. This pioneering project aims to enhance the accuracy and efficiency of surgical processes through cutting-edge AI technology.

Revolutionary Advancements in Medical Video Analysis

The uAI NEXUS MedVLM is set to significantly transform the way medical videos are interpreted, highlighting its exceptional spatial and temporal precision tailored to clinical environments. Developed on a monumental dataset of 531,850 video-instruction pairs across eight clinical settings—including robotic surgery, laparoscopic procedures, endoscopy, and open surgery—the model showcases outstanding performance, well beyond existing general-purpose foundations like GPT-5.4 and Gemini 3.1.

Highlighting its capabilities, the uAI NEXUS achieves an astounding 89.4% accuracy in evaluating surgical safety, leaving behind its competitors significantly (GPT-5.4 at 1.8% and Gemini 3.1 at 10.1%). Furthermore, it excels in spatial-temporal action localization, demonstrating up to 14 times higher mIoU scores compared to GPT-5.4, solidifying its reputation as the leading model for medical video analytics.

Setting New Standards with Open Data Resources

As part of its commitment to collaborative improvement, UII is initiating the global open challenge to accelerate innovation in the development of LLM models for medical video analysis. This endeavor commences with the phased release of its meticulously crafted dataset “MedVidBench,” featuring 6,245 rigorously vetted benchmark test samples. This dataset is set to foster a new level of clinical precision, enabling developers to evaluate their models through a standardized ranking system based on private reference data, ensuring continuous updates in global performance assessments.

The initiative not only enhances the competitive landscape but also invites AI researchers, developers, and healthcare institutions worldwide to join in contributing towards the advancement of medical video intelligence.

Overcoming Traditional Constraints in AI

The understanding and analysis of medical videos have long posed significant challenges within the realm of artificial intelligence, mainly due to the demand for precise spatial awareness, intricate temporal logic, and unwavering clinical accuracy. Traditionally, these advancements have been impeded by the chronic scarcity of clinical data and exorbitant costs associated with expert annotations.

With the introduction of uAI NEXUS MedVLM, UII aims to eliminate these bottlenecks. By developing an extensive frame-by-frame annotation framework for diverse clinical videos, critical attributes such as instrument paths, spatial positioning, precise surgical maneuvers, and vital risk indicators have been rigorously captured. This comprehensive database empowers uAI NEXUS MedVLM with a robust clinical intelligence stack, marking its utility in the medical field.

Clinical Integration and Future Prospects

Designed specifically for clinical implementations, the uAI NEXUS MedVLM facilitates informed decision-making and data-driven quality control across surgical workflows, simultaneously shortening the learning curve for clinicians while improving training efficiency and consistency. Going forward, the model has the potential to serve as a core perception and cognition engine for embodied AI, laying the groundwork for more automated, standardized, and intelligent healthcare ecosystems.

In conclusion, the launch of uAI NEXUS MedVLM by United Imaging Intelligence not only represents a significant milestone in medical video interpretation but also paves the way for collaborative innovations in AI-driven healthcare solutions and practices.

For more details, visit the project page.