RoboAbstention - Ambiguous Referent: leaderboard

The Ambiguous Referent category of RoboAbstention (2026, The Yes-Man Syndrome): 420 instructions whose target matches several objects in 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 420 Ambiguous 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 Flash6#463
2Llama 4 Maverick3.8#1110
3Claude Sonnet 4.62.4#173
4Gemini Robotics-ER 1.61.9#170
5GPT-5.4 MiniHigh1.7#406
6Qwen 3.5 27B1.4#375
7GPT-5.4 Nano0.7#651
8GPT-5.40#135

Interactive version: theaggregate.ai/benchmark?slug=roboabstention-ambiguous-referent · How It Works · Data refreshed daily, snapshot 2026-11-21.