Ant Group Open-Sources 124B Medical Multimodal Model LingSante-3.1-air-VL
Ant Group has open-sourced LingSante-3.1-air-VL, a medical multimodal model built on a mixture-of-experts architecture with 124B total and 5.5B active parameters per token, covering medical image understanding, report generation and medical reasoning. Ant reports it leads GPT-6.1 Sol, Claude Opus 5.5 and Gemini 3.8 Flash across AFUMED-Derm, AFUMED-Report, MIMIC-CXR and medical VQA, while remaining competitive on general tasks such as BrowseComp-VL and AndroidWorld. Its reinforcement learning rewards clinical meaning, abnormality localization, false positives and missed findings rather than report style imitation.
Full Report
Rohan Paul (@rohanpaul_ai) · 2026-11-22 21
Ant Group just open-sourced LingSante-3.1-air-VL, a medical model that writes chest X-ray reports.
beats GPT-6.1 Sol, Claude Opus 5.5 and Gemini 3.8 Flash on all 3 standard report benchmarks.
its a mixture-of-experts (MoE) model with 124B total parameters, of which only 5.5B are active per token
its reinforcement learning scored each X-ray report on clinical meaning, abnormality location, false positives and missed findings, which rewards a correct diagnosis rather than only copying the style of example reports.
↳ 引用 @AntLingAGI (Ant Ling): Meet LingSante-3.1-air-VL, an open-source multimodal medical model with 124B total and 5.5B active parameters. It combines medical image understanding, report generation and medical reasoning with multimodal search and tool use. In our evaluation, it leads the compared models on AFUMED-Derm, AFUMED-Report, MIMIC-CXR and medical VQA (averaged across benchmarks), ahead of GPT-6.1 Sol, Gemini 3.8 Flash on all four. It also demonstrates strong general-purpose capabilities, scoring higher than GPT-6.1 Sol on BrowseComp-VL and Gemini 3.8 Flash on AndroidWorld, alongside competitive performance on MMSearch. Explore LingSante-3.1-air-VL and build with it.
