RoboAbstention - Underspecified Intent: leaderboard

The Underspecified Intent category of RoboAbstention (2026, The Yes-Man Syndrome): 1246 instructions that leave out what the robot must know to act, 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 1246 Underspecified Intent 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. Saturation forecast: Around 2029. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Sonnet 4.623.7#173
2Gemini 2.5 Flash20.5#463
3GPT-5.4 Nano19.6#651
4GPT-5.419.3#135
5GPT-5.4 MiniNon-reasoning17.7#406
6Llama 4 Maverick4#1110
7Gemini Robotics-ER 1.62.6#170
8Qwen 3.5 27B1.8#375

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

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