UTQA: Undergraduate Thermodynamics Question Answering Benchmark

UTQA is a 50-item multiple-choice benchmark designed to evaluate large language models on undergraduate-level thermodynamics.
It consists of single-choice questions (four options, one correct), of which 17 include associated figures.
The benchmark targets reasoning about state functions, reversibility, entropy, and diagram interpretation, rather than simple plug-and-chug calculations.

---

Dataset Structure

• Number of items: 50
• Format: CSV table + images folder
• Columns:
  • question_number: integer identifier
  • question: question stem (text)
  • option_a – option_d: four answer choices
  • image: relative path to figure (if applicable; empty otherwise)
  • correct_answer: one of {a, b, c, d}
  • explanation: text explanation of the solution
  • image_explain: additional explanation of the solution using a figure (if applicable)

Images are provided in the /images/ directory and linked from the image column.

---

Intended Use

UTQA is released as a benchmark for evaluation of language models and as a resource for educators and researchers studying AI capabilities in unsupervised teaching contexts.
It is not intended for use in pretraining or fine-tuning language models.

✅ Allowed

• Model evaluation and benchmarking
• Academic research
• Teaching and educational analysis
• Publication of results with proper citation

❌ Not Allowed

• Pretraining or fine-tuning of models on UTQA (in whole or in part)
• Incorporation into larger training corpora
• Commercial use for training or data augmentation

---

License

This dataset is released under a Custom Evaluation-Only License (see LICENSE).

• Free to use for research, education, and evaluation.
• Training use is prohibited.
• Attribution is required in all uses and publications.

---

Citation

If you use UTQA in your research, please cite:

Geißler, A., Bien, L.-S., Schöppler, F., & Hertel, T. (2025).
From Canonical to Complex: Benchmarking LLM Capabilities in Undergraduate Thermodynamics.
arXiv:2508.21452 [physics.ed-ph]. https://arxiv.org/abs/2508.21452

@misc{Geissler2025,
  title        = {From Canonical to Complex: Benchmarking LLM Capabilities in Undergraduate Thermodynamics}, 
  author       = {Anna Geißler and Luca-Sophie Bien and Friedrich Schöppler and Tobias Hertel},
  year         = {2025},
  eprint       = {2508.21452},
  archivePrefix= {arXiv},
  primaryClass = {physics.ed-ph},
  url          = {https://arxiv.org/abs/2508.21452},
}