LaViSA - Verb Phrase Attachment: leaderboard
Metric: Per-trial accuracy (%) on the verb phrase attachment 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 50; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 97 |
| 2 | GPT-5.2 | 95 |
| 3 | Gemini 3.1 Flash Lite | 93.5 |
| 4 | Qwen 3 VL 32B (Thinking) | 92.5 |
| 5 | Qwen 3 VL 32B Instruct | 86 |
| 6 | Qwen 3 VL 8B (Thinking) | 86 |
| 7 | Qwen 3 VL 4B Instruct | 82.5 |
| 8 | Qwen 3 VL 4B (Thinking) | 79 |
| 9 | Qwen 3 VL 8B Instruct | 71 |
| 10 | Gemma 3 12B (IT) | 67.5 |
| 11 | Gemma 3 27B (IT) | 65 |
| 12 | Gemma 3 4B (IT) | 52.5 |
Interactive version: theaggregate.ai/benchmark?slug=lavisa-verb-phrase-attachment · How It Works · Data refreshed daily, snapshot 2026-09-29.