RoboAbstention - Contradictory Instructions: leaderboard

The Contradictory Instructions category of RoboAbstention (2026, The Yes-Man Syndrome): 837 instructions that contradict themselves, given with an image to a vision-language robot planner that should abstain or ask. The score is the abstention rate in percent; higher is better.

Metric: Abstention rate (%) on the 837 Contradictory Instructions instructions, image plus instruction to an embodied VLM planner, every instruction warranting abstention, act or abstain judged by GPT-5.4 mini (97.5 percent agreement with human labels), API default sampling; higher is better. Source: arxiv.org. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 2.5 Flash36.3#466
2GPT-5.429#138
3Llama 4 Maverick28.6#1113
4Claude Sonnet 4.626.5#179
5GPT-5.4 Nano22.6#653
6GPT-5.4 MiniLow22#409
7Gemini Robotics-ER 1.616.6#175
8Qwen 3.5 27B5.4#380

Interactive version: theaggregate.ai/benchmark?slug=roboabstention-contradictory-instructions · How It Works · Data refreshed daily, snapshot 2026-11-22.