RS-Neg MCQ - Object Negation: leaderboard
Choosing the one correct caption of four, where the distractors mix affirmative, negated and hybrid descriptions, restricted to the items whose negation targets an object. RS-Neg (Han et al., Peng Cheng Laboratory, 2026) tests negation understanding on remote-sensing images, with negated queries written by LLMs from image captions and checked against the image by a detector-guided visual search. Accuracy in percent over 4,502 items, answered zero-shot; higher is better.
Metric: Accuracy (%; pick the one correct caption of four, the distractors mixing affirmative, negated and hybrid descriptions; the 4,502 items whose negation is object-level; zero-shot). Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 57.29 | #982 |
| 2 | Qwen 2.5 VL 7B Instruct | 54.22 | #1842 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rs-neg-mcq-object-negation · How It Works · Data refreshed daily, snapshot 2026-11-12.