The ELFVET Multimodal Veterinary Medical Large Model achieves ≥92% accuracy in lesion recognition for dermatological conditions, ear canal parasites, and oral issues; AI-generated diagnostic recommendations and treatment plans achieve ≥96% accuracy. It is important to emphasize that the system serves as an AI-assisted decision-support tool — an "AI Co-Brain" for veterinarians — with final diagnosis and treatment decisions remaining the responsibility of licensed veterinarians; AI does not replace clinical judgment. Our accuracy is supported on four levels: first, the model is built on the industry-first Mixture-of-Experts (MoE) architecture that combines a generalist brain with specialist sub-models to reduce hallucinations in general-purpose large models, with ongoing advances for dermatology, ear canal, and heart sound analysis; second, the core training set comprises millions of precisely annotated data points together with the Chinese Veterinary Pharmacopoeia database; third, all veterinary medical data and cases are reviewed by veterinary experts; and fourth, the models powering the product have been deployed and validated through stress testing and multiple iterative cycles across hundreds of pet hospitals and stores.

The Pet Veterinarian AI Diagnosis and Treatment Assistant provides AI-assisted imaging analysis. The model can perform preliminary interpretation of macroscopic imaging such as X-rays and CT scans, identifying common anatomical abnormalities, organ lesions, and skeletal injuries to give veterinarians broader decision support. AI-based image interpretation serves as a clinical auxiliary reference, with final conclusions confirmed by the veterinarian in conjunction with clinical findings.

The portable ultrasound probe bundled with the Pet Veterinarian AI Diagnosis and Treatment Assistant offers the following specifications: 256 grayscale levels with support for broadband, panoramic focus, and harmonic imaging; probe frequency of 2–12 MHz; and a maximum scan depth of 310 mm. Wirelessly integrated via a lightweight Pad hub, the probe supports mobile clinical scenarios such as house calls, inpatient rounds, and exotic-pet care, forming a core component of the hospital's portable diagnostic toolkit.

The ELFVET large model is built on four traceable data sources: first, first-hand multimodal data — covering skin, ear canal, oral cavity, heart sounds, ECG and more — collected via proprietary intelligent hardware such as the 500x microscope and AI Smart Stethoscope, which has been curated into a training set of millions of precisely annotated records; second, integration with the Chinese Veterinary Pharmacopoeia and the registered veterinary drug database; third, millions of professional clinical cases combined with pharmacopoeia data; and fourth, industry-academia-research collaboration with leading agricultural universities in China, with specialist experts involved in data annotation and review. All data is fed back into the ELFVET large model for training and continuous iteration.