Hugging Face's logo

Datasets:
launch
/
LudoBench

Tasks:
Visual Question Answering
Question Answering
Modalities:
Image
Tabular
Text
Formats:
parquet
Languages:
English
Size:
< 1K
Tags:
board-games
multimodal-reasoning
benchmark
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
Community
1
LudoBench
File size: 2,533 Bytes
dd23d71
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
 
---
dataset_info:
  features:
    - name: ID
      dtype: int64
    - name: Game
      dtype: string
    - name: tier
      dtype: int64
    - name: Question
      dtype: string
    - name: Answer
      dtype: string
    - name: game_state_url
      dtype: string
configs:
  - config_name: default
    data_files:
      - split: test
        path: ludobench.parquet
license: mit
task_categories:
  - visual-question-answering
  - question-answering
language:
  - en
size_categories:
  - n<1K
pretty_name: LudoBench
tags:
  - board-games
  - multimodal-reasoning
  - benchmark
---

# LudoBench

**LLMs as Rules Oracles: Exploring Real-World Multimodal Reasoning in Tabletop Strategy Game Environments**

*ICLR 2026*

[![Paper](https://img.shields.io/badge/Paper-OpenReview-blue)](https://openreview.net/forum?id=TOgQ00DEek)
[![Dataset](https://img.shields.io/badge/Dataset-HuggingFace-yellow)](https://huggingface.co/datasets/launch/LudoBench)
[![GitHub](https://img.shields.io/badge/Code-GitHub-black)](https://github.com/jpeper/LudoBench)
[![Demo](https://img.shields.io/badge/Demo-HF%20Space-orange)](https://huggingface.co/spaces/launch/LudoBench)

---

A multimodal board-game reasoning benchmark evaluating LLM/VLM reasoning
across 5 strategy games and 3 difficulty tiers.

## Dataset Description

- **638** annotated question-answer pairs
- **5 games**: Kingdomino, Res Arcana, Pax Renaissance, Carcassonne, Catan
- **3 tiers**: Environment Perception (T1), Rules Integration (T2), Short-Horizon Optimization (T3)
- **3 rules modalities**: None (parametric), Text (text rulebook), Image (image rulebook)

## Fields

| Field | Description |
|-------|-------------|
| `ID` | Unique question identifier |
| `Game` | Board game name |
| `tier` | Difficulty tier (1, 2, or 3) |
| `Question` | The question text |
| `Answer` | Expected answer |
| `game_state_url` | Path(s) to game state image(s), semicolon-separated if multiple |

## Benchmark Results

See `benchmark_results.csv` for accuracy scores of 9 models across all game/tier/modality splits.

## Citation

```bibtex
@inproceedings{peper2026ludobench,
  title={{LLMs} as Rules Oracles: Exploring Real-World Multimodal Reasoning in Tabletop Strategy Game Environments},
  author={Peper, Joseph J. and Gandra, Sai Krishna and Zhang, Yunxiang and Chennareddy, Vaibhav and Jha, Shloki and Payani, Ali and Wang, Lu},
  booktitle={Proceedings of the Fourteenth International Conference on Learning Representations (ICLR)},
  year={2026},
  address={Rio de Janeiro, Brazil}
}
```