ReaLM - FOA-Attack: leaderboard

Metric: Attack success rate (%; share of adversarial cases where the model gives an incorrect, task-relevant answer) under FOA-Attack (optimal-transport feature matching against a CLIP surrogate ensemble, L-inf 16/255) on the 832 ReaLM physical-world cases (driving, manipulation, grasping, physics, egocentric, diagnostics and scene QA; image, question and ground-truth answer), black-box one-shot evaluation at temperature 0, a Qwen3-8B judge extracting the answer; lower is better. Source: arxiv.org. Saturation forecast: Around 2029. 13 models tracked.

Top models

#ModelScore
1Gemini 3 Flash45.4
2Qwen 3.5 27B45.8
3Seed 2.0 Mini46.9
4Qwen 3 VL 32B47.1
5Qwen 3.5 122B A10B47.2
6Qwen 3.6 Flash47.2
7GPT-4.1 Mini47.4
8Kimi K2.548
9Qwen 3.5 9B48.3

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