During December 2025, Claas developed a Vision-Language Model Visual Reasoning Benchmark for the EvolvingLMMs-Lab/lmms-eval repository, focusing on quantifying basic visual reasoning abilities in machine learning models. Claas designed and implemented a reproducible workflow in Python and YAML, incorporating clear evaluation metrics and comprehensive documentation. The benchmark uses simple geometric tasks to systematically reveal perceptual limitations in current vision-language models, providing researchers with a practical tool to measure progress and guide future improvements. By integrating the benchmark end-to-end within the repository, Claas enabled streamlined experimentation and comparison, demonstrating depth in benchmarking, data analysis, and machine learning engineering practices.