Quantifying Multilingual Performance of Large Language Models Across Languages