RoboAbstention - Missing Referent: leaderboard

The Missing Referent category of RoboAbstention (2026, The Yes-Man Syndrome): 1250 instructions that name an object absent from the scene, 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 1250 Missing Referent 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 Flash75.2#466
2Claude Sonnet 4.659.4#179
3Llama 4 Maverick47.7#1113
4Gemini Robotics-ER 1.631.7#175
5Qwen 3.5 27B23.2#380
6GPT-5.4 MiniNon-reasoning12.7#409
7GPT-5.410.4#138
8GPT-5.4 Nano9.8#653

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