RoboAbstention - Missing Capability: leaderboard
The Missing Capability category of RoboAbstention (2026, The Yes-Man Syndrome): 1067 instructions that need a capability the robot lacks, 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 1067 Missing Capability 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
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 2.5 Flash | 44.8 | #463 | |
| 2 | GPT-5.4 Mini | Low | 40.9 | #406 |
| 3 | Gemini Robotics-ER 1.6 | 19.6 | #170 | |
| 4 | Claude Sonnet 4.6 | 17.9 | #173 | |
| 5 | GPT-5.4 Nano | 16.5 | #651 | |
| 6 | Qwen 3.5 27B | 9.5 | #375 | |
| 7 | GPT-5.4 | 8.5 | #135 | |
| 8 | Llama 4 Maverick | 7.2 | #1110 |
Interactive version: theaggregate.ai/benchmark?slug=roboabstention-missing-capability · How It Works · Data refreshed daily, snapshot 2026-11-21.