PriVE-Bench - Medical Modality: leaderboard

Metric: Counterfactual accuracy (%) in the medical modality domain; paired original and counterfactual images in five domains, counterfactual-only input, answers classified by a text-only judge (GPT-4o-mini for closed models, Qwen3-8B for open models); temperature 0, reasoning off or minimal. Source: arxiv.org. Saturation forecast: Around 2033. 8 models tracked.

Top models

#ModelScore
1Gemma 3 12B (IT)66.8
2Claude Sonnet 441.5
3Qwen 3 VL 32B Instruct33.1
4GPT-5 (Minimal)31.5
5Qwen 3 VL 8B Instruct27.2
6InternVL3.5-8B19.9
7Gemma 3 27B (IT)19.2

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