RebusBench (1-Shot): leaderboard
RebusBench (2026) tests vision-language models on 1,164 human-authored rebus puzzles, where the picture and its text layout are clues to an idiom or phrase. Exact match (%) after normalization, with one solved example in the prompt; higher is better.
Metric: Exact match: exact match accuracy (%) against the gold answer after lowercasing and stripping whitespace and special characters, on the 1,164 human-authored rebus puzzles of RebusBench, one in-context example in the text prompt; higher is better. Source: arxiv.org. Saturation forecast: Around February 2028. 9 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 32B Instruct | 7.39 | #1062 |
| 2 | Qwen 3 VL 8B | 6.36 | #917 |
| 3 | Qwen 2.5 VL 7B Instruct | 5.15 | #1902 |
| 4 | InternVL3.5-30B-A3B | 5.15 | #1267 |
| 5 | InternVL3.5-8B | 4.81 | #1604 |
| 6 | InternVL3.5-4B | 4.73 | #1930 |
| 7 | Qwen 3 VL 4B | 4.12 | #1151 |
| 8 | Qwen 2.5 VL 3B Instruct | 3.52 | #2713 |
| 9 | LLaVA-v1.5-7B | 1.2 | #7630 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=rebusbench-1-shot · How It Works · Data refreshed daily, snapshot 2026-11-21.