ClinHallu: leaderboard

Metric: Answer accuracy (%), macro-averaged over the ClinHallu test items of VQA-RAD, PathVQA, MedFrameQA and MedXpertQA (medical VQA with validated reference traces of visual recognition, knowledge recall and reasoning integration); the model writes a structured trace and answer (temperature 0.7) and a Qwen3.5-27B judge grades them (temperature 0.01); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 11 models tracked.

Top models

#ModelScore
1Gemini 3 Flash80.1
2Qwen 3.5 9B69.1
3Qwen 3.5 4B64.3
4Qwen 3 VL 32B63.8
5InternVL3.5-8B53.9
6MedGemma-4B-IT53.2
7Lingshu-7B52.7
8Qwen 2.5 VL 7B Instruct42.7

Interactive version: theaggregate.ai/benchmark?slug=clinhallu · How It Works · Data refreshed daily, snapshot 2026-09-29.