AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Lingshu: Generalist Foundation Model for Unified Multimodal Medical Understanding and Reasoning

Weiwen Xu, Hou Pong Chan, Long Li, Mahani Aljunied, Ruifeng Yuan, Jianyu Wang, Chenghao Xiao, 陈贵珍, Chaoqun Liu, Zhaodonghui Li, Yu Sun, Junao Shen and 8 more

TPAMI 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Multimodal LLM benchmarks · 319 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add medical text data (173K samples) to instruction tuning (ablation)Lingshu-Med-7B · MMMU-Medaccuracy: 50.9 -> 54.0Improved
add medical text data to instruction tuning (ablation)Lingshu-Med-7B · MedXpertQAaccuracy: 24.0 -> 26.7Improved
add medical multimodal data (2.7M) to instruction tuning (ablation)Lingshu-Med-7B · SLAKEaccuracy: 69.0 -> 83.1Improved
add medical multimodal data to instruction tuning (ablation)Lingshu-Med-7B · PathVQAaccuracy: 45.2 -> 61.9Improved
add general multimodal data (1.2M) to instruction tuning (ablation)Lingshu-Med-7B · MMMU-Medaccuracy: 51.4 -> 54.0Improved
add general text data (1M) to instruction tuning (ablation)Lingshu-Med-7B · VQA-RADaccuracy: 65.2 -> 67.9Improved
add synthetic multimodal medical data (1.1M) (ablation)Lingshu-Med-7B · OmniMedVQAaccuracy: 71.6 -> 82.9Improved
add distilled medical CoT text (162K) (ablation)Lingshu-Med-7B · PMC-VQAaccuracy: 53.6 -> 56.3Improved
add Stage 3 medical captions (2M) (ablation)Lingshu-Med-7B · MMMU-Medaccuracy: 52.9 -> 54.0Improved
add Stage 1&2 caption alignment (4.8M captions) (ablation)Lingshu-Med-7B · PMC-VQAaccuracy: 54.9 -> 56.3Improved
add alignment stages before Stage 3 vs direct fine-tuning of Qwen2.5-VL on Stage 3 data (ablation)Lingshu-Med-7B · medical benchmarks averageaccuracyImproved
apply RLVR (GRPO) on top of Lingshu-Med-7BLingshu-Med-7B · medical multimodal avg (7 benchmarks)accuracy: 61.8 -> 61.5No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy