LaViSA - Conjunction Scope: leaderboard
Metric: Per-trial accuracy (%) on the conjunction scope category; task: pick which disambiguated reading of a structurally ambiguous sentence a generated image depicts (2 or 3 readings per sentence, 100 sentences per category); chance 33.3; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 85 |
| 2 | Qwen 3 VL 32B Instruct | 78.3 |
| 3 | Gemini 3.1 Flash Lite | 76.3 |
| 4 | Qwen 3 VL 4B Instruct | 73.7 |
| 5 | GPT-5.2 | 73 |
| 6 | Gemma 3 27B (IT) | 67.3 |
| 7 | Qwen 3 VL 32B (Thinking) | 62.3 |
| 8 | Gemma 3 12B (IT) | 59.7 |
| 9 | Qwen 3 VL 8B Instruct | 51.7 |
| 10 | Qwen 3 VL 8B (Thinking) | 46.3 |
| 11 | Qwen 3 VL 4B (Thinking) | 43.3 |
| 12 | Gemma 3 4B (IT) | 38 |
Interactive version: theaggregate.ai/benchmark?slug=lavisa-conjunction-scope · How It Works · Data refreshed daily, snapshot 2026-09-29.