---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: exam
    dtype: string
  - name: topic
    dtype: string
  - name: sinav
    dtype: string
  - name: konu
    dtype: string
  - name: answer
    dtype: string
  splits:
  - name: train
    num_bytes: 168030625.25
    num_examples: 1854
  download_size: 167855901
  dataset_size: 168030625.25
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: mit
task_categories:
- image-text-to-text
- image-to-text
language:
- tr
- en
pretty_name: YKS - Uniform
size_categories:
- 1K<n<10K
---
# YKS Uniform

<!-- Provide a quick summary of the dataset. -->

![Scientia dux vitae certissimus](https://yks-uniform.github.io/assets/bilim_agaci.jpg)

Large Language Models (LLMs) and Large Multimodal Models (LMMs) demonstrate impressive problem-solving skills across many tasks and domains. However, their ability to reason over structured, curriculum-based educational questions—particularly in the context of Turkish high school entrance examinations—has not been systematically studied.

To address this gap, we introduce YKS Uniform, a balanced multimodal benchmark covering the Turkish high school curriculum with equal representation across all topics. By sampling six questions per topic, we constructed a dataset of 1,854 multimodal questions spanning both TYT and AYT exams. These questions require deep reasoning over text, diagrams, and exam-style contexts.

Using this benchmark, we conducted a comprehensive evaluation of 16 open-weight and proprietary models. Our results highlight both the strengths and limitations of current models in handling exam-style reasoning tasks. The best-performing system, Gemini-2.5-Flash, achieved an overall accuracy of %84.7, substantially higher than open-weight alternatives but still leaving a measurable gap to human-level performance.

Visit [YKS Uniform](https://yks-uniform.github.io/) for details.

## Dataset Details

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->
Developed by METU Computer Engineering. A 1854 multimodal exam preparation questions spanning 309 topics in Turkish High School curriculum. Each topic is equally represented by sampling first six questions of each topic's test from publicly available e-books of a publisher (hence the name Uniform). For academic use only. 


- **Curated by:** Egemen Sert, Şeyda Ertekin
- **Language(s) (NLP):** Turkish, English
- **License:** MIT

## Uses

<!-- Address questions around how the dataset is intended to be used. -->

This is a benchmark dataset for academic use only. Please use the dataset exclusively for benchmarking purposes. 

## Citation

<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->

**BibTeX:**

```
@misc{yksuniform2025,
  title        = {YKS Uniform: A Balanced Multimodal Benchmark Covering the Turkish High School Curriculum},
  author       = {Sert, Egemen and Ertekin, Şeyda},
  year         = {2025},
  howpublished = {\url{https://yks-uniform.github.io/}},
  note         = {Accessed: 2025-08-23}
}
```

## Dataset Card Contact

Please contact egemen.sert@metu.edu.tr for submissions and benchmark related questions.