ClinHallu - Visual Recognition Hallucination: leaderboard

Metric: Visual recognition hallucination rate (%): share of items whose generated visual-recognition stage the judge marks as hallucinated against the reference, 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); lower is better. Source: arxiv.org. Saturation forecast: Around December 2026. 11 models tracked.

Top models

#ModelScore
1Gemini 3 Flash25.8
2Qwen 3.5 9B41.9
3InternVL3.5-8B45.6
4Qwen 3 VL 32B50.8
5MedGemma-4B-IT51.1
6Qwen 3.5 4B52
7Lingshu-7B52.2
8Qwen 2.5 VL 7B Instruct65.9

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