Lingshu-32B: benchmark results
Provider: Alibaba. Released 2025-06-05. Access: Open.
Aggregate Rating 1588 ± 13, rank #1141 of 10916 rated models, from 142 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LoMeVQA - Progress Classification | 54.8 | Accuracy (%; three-option classification of disease progress | 100 |
| LoMeVQA - Progress Description | 26.67 | F1-RadGraph clinical efficacy (%; free-text description of t | 100 |
| LoMeVQA - Progress Report Generation | 16.45 | F1-RadGraph clinical efficacy (%; progress report written fr | 100 |
| MMBU - Grounded Classification (Boxes, Open-Ended) | 15 | Open-ended classification of a region marked by a bounding b | 100 |
| Medical MM Leaderboard - Average | 66.6 | Average Accuracy (%) | 100 |
| Medical MM Leaderboard - OMVQA | 83.4 | Accuracy (%) | 100 |
| Medical MM Leaderboard - PathVQA | 65.9 | Accuracy (%) | 100 |
| Medical MM Leaderboard - SLAKE | 89.2 | Accuracy (%) | 100 |
| Medical MM Leaderboard - VQA-RAD | 76.5 | Accuracy (%) | 100 |
| Seizure-Semiology-Suite - Report Generation | 39.8 | Seizure Report Quality Index, Seizure RQI (0-100): weighted | 100 |
| ADRD-Bench - Caregiving QA | 95.3 | Exact-match accuracy (0-1, shown times 100) on 149 caregivin | 95.6 |
| Pocket-Dentist - BRAR Bone Resorption Grading | 39.2 | Macro-F1 (%; three bone-resorption grades from a panoramic r | 92.3 |
Interactive version: theaggregate.ai/model?slug=lingshu-32b · How It Works · Data refreshed daily, snapshot 2026-11-08.