LaViSA - Ellipsis: leaderboard
Metric: Per-trial accuracy (%) on the ellipsis 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: Around December 2026. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 83.2 |
| 2 | GPT-5.2 | 71.1 |
| 3 | Gemini 3.1 Flash Lite | 70.8 |
| 4 | Qwen 3 VL 8B Instruct | 66.8 |
| 5 | Qwen 3 VL 32B (Thinking) | 65.8 |
| 6 | Qwen 3 VL 32B Instruct | 63.9 |
| 7 | Gemma 3 12B (IT) | 61.9 |
| 8 | Gemma 3 27B (IT) | 61.4 |
| 9 | Qwen 3 VL 4B Instruct | 61.4 |
| 10 | Qwen 3 VL 8B (Thinking) | 59.4 |
| 11 | Gemma 3 4B (IT) | 54.5 |
| 12 | Qwen 3 VL 4B (Thinking) | 51 |
Interactive version: theaggregate.ai/benchmark?slug=lavisa-ellipsis · How It Works · Data refreshed daily, snapshot 2026-09-29.