Ant Group's Bailing Medical Model Sante: With only 5.1 billion activated parameters, it outperforms GPT-5.6 Sol in diagnostic tasks.

Insight from Beating AI News: Ant Group’s Lingbao AI has launched the medical health model Ling-3.0-flash-Sante. Built on Ling-3.0-flash with enhanced medical capabilities, the model uses a Mixture of Experts (MoE) architecture, with 124 billion total parameters, only ~5.1 billion of which are activated per token. Key capabilities include medical reasoning, drug safety assessment, evidence-based retrieval, and long-process medical research. In benchmarks released by Lingbao, Sante scored 83.8 on DiagnosisArena-MCQ, outperforming GPT-5.6 Sol (81.9), Kimi K3 (78.4), and Gemini 3.6 Flash (76.2). It notched a 53.9 on MedXpertQA-Text, slightly higher than Kimi K3’s 53.5, though still below GPT-5.6 Sol’s 60.2 and Gemini 3.6 Flash’s 62.4. Notably, the model isn’t limited to excelling at medical multiple-choice questions. Lingbao also tested Sante on BrowseComp, DeepSearchQA, and HLE with tools, focusing on strengthening its ability to "retrieve information independently before answering". Separate tests for medical ethics and safety returned a score of 82.1 on MedEthicAlign and 78.6 on internal security test AFUSAFE-MedSCE. Sante is now integrated with OpenRouter and Vercel, available for free access.