![[2602.19517] Classroom Final Exam: An Instructor-Tested Reasoning Benchmark](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19517] Classroom Final Exam: An Instructor-Tested Reasoning Benchmark
Summary
The paper presents CFE, a multimodal benchmark for evaluating large language models' reasoning capabilities in STEM domains, highlighting challenges faced by current models.
Why It Matters
This research is significant as it addresses the limitations of large language models in reasoning tasks, providing a structured benchmark that can guide future improvements in AI capabilities across various STEM fields. It emphasizes the need for better reasoning efficiency and accuracy in AI systems.
Key Takeaways
- Introduction of CFE, a benchmark for reasoning in STEM.
- Current models struggle with maintaining correct intermediate states in multi-step solutions.
- The benchmark reveals significant room for improvement in model accuracy.
- Model-generated solutions often involve more reasoning steps, increasing error risk.
- Data and code for the benchmark are publicly available for further research.
Computer Science > Artificial Intelligence arXiv:2602.19517 (cs) [Submitted on 23 Feb 2026] Title:Classroom Final Exam: An Instructor-Tested Reasoning Benchmark Authors:Chongyang Gao, Diji Yang, Shuyan Zhou, Xichen Yan, Luchuan Song, Shuo Li, Kezhen Chen View a PDF of the paper titled Classroom Final Exam: An Instructor-Tested Reasoning Benchmark, by Chongyang Gao and Diji Yang and Shuyan Zhou and Xichen Yan and Luchuan Song and Shuo Li and Kezhen Chen View PDF HTML (experimental) Abstract:We introduce \CFE{} (\textbf{C}lassroom \textbf{F}inal \textbf{E}xam), a multimodal benchmark for evaluating the reasoning capabilities of large language models across more than 20 STEM domains. \CFE{} is curated from repeatedly used, authentic university homework and exam problems, together with reference solutions provided by course instructors. \CFE{} presents a significant challenge even for frontier models: the newly released Gemini-3.1-pro-preview achieves an overall accuracy of 59.69\%, while the second-best model, Gemini-3-flash-preview, reaches 55.46\%, leaving considerable room for improvement. Beyond leaderboard results, we perform a diagnostic analysis by decomposing reference solutions into reasoning flows. We find that although frontier models can often answer intermediate sub-questions correctly, they struggle to reliably derive and maintain correct intermediate states throughout multi-step solutions. We further observe that model-generated solutions typically have more re...