This research evaluates 26 Vision Language Models on their understanding of core mechanical engineering concepts through 155 cognitive experiments.

• Models were tested on system stability, gear systems, pulleys, leverage, and fluid mechanics
• Reveals current capabilities and limitations in AI's understanding of physical principles
• Establishes a benchmark for measuring mechanical reasoning in AI systems
• Identifies gaps between human and AI understanding of fundamental physics

For engineering teams, this research provides critical insights into how AI systems can potentially assist with mechanical design tasks and where human expertise remains essential.

Probing Mechanical Reasoning in Large Vision Language Models