WYRipple/S1-Bench

The benchmark constructed in paper S1-Bench: A Simple Benchmark for Evaluating System 1 Thinking Capability of Large Reasoning Models.

Introduction

S1-Bench is a novel benchmark designed to evaluate Large Reasoning Models' performance on simple tasks that favor intuitive system 1 thinking rather than deliberative system 2 reasoning. S1-Bench comprises 422 question-answer pairs across four major categories and 28 subcategories, balanced with 220 English and 202 Chinese questions.

Meta Data

The data set has

• ID: a int feature.
• question: a string feature.
• answer: a string feature.
• language: a string feature.
• category: a string feature.
• sub_category: a string feature.

Example

{
    "ID": 204,
    "question": "What's 7 plus 8?",
    "answer": "15",
    "language": "en",
    "category": "reasoning_question",
    "sub_category": "numerical_reasoning"
}

{
    "ID": 156,
    "question": "单词"hola"是西班牙语还是法语？",
    "answer": "西班牙语",
    "language": "zh",
    "category": "analysis_questions",
    "sub_category": "language_classification"
}

{
    "ID": 105,
    "question": "输出一个字母“Z”",
    "answer": "Z",
    "language": "en",
    "category": "instruction_following",
    "sub_category": "repetition_constraints"
}

{
    "ID": 294,
    "question": "How many minutes equal 1 hour?",
    "answer": "60",
    "language": "en",
    "category": "knowledge_question",
    "sub_category": "measurement_units"
}

Citation

If you find our work useful, please consider citing our paper:

@misc{zhang2025s1benchsimplebenchmarkevaluating,
      title={S1-Bench: A Simple Benchmark for Evaluating System 1 Thinking Capability of Large Reasoning Models}, 
      author={Wenyuan Zhang and Shuaiyi Nie and Xinghua Zhang and Zefeng Zhang and Tingwen Liu},
      year={2025},
      eprint={2504.10368},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2504.10368}, 
}