This study assesses how effectively Large Language Models can solve complex thermodynamic problems, creating a new benchmark for AI capabilities in engineering domains.

• Tested 5 leading LLMs (GPT-3.5, GPT-4, GPT-4o, Llama 3.1, MistralAI) on 22 thermodynamic problems
• Developed a comprehensive benchmark covering both basic and advanced thermodynamic concepts
• Measured specific engineering problem-solving capabilities rather than general knowledge
• Provides insights into which models are most reliable for technical engineering applications

This research has significant implications for engineering education and professional practice, potentially transforming how AI tools can support complex technical problem-solving in specialized fields.

Using Large Language Models for Solving Thermodynamic Problems