Comparative Analysis of ChatGPT-4o and DeepSeek-R1 for Multilingual NLP Tasks

The Recent advancements in AI technology have become an inseparable part of many industries worldwide which necessitates the availability and reliability of inclusive AI models for multilingual environments. The majority of AI models are predominantly trained using high-resource languages such as English while there is less emphasis on lowresource languages such as Indonesian with studies showing a great bias towards English inputs yielding more favorable results than non-English languages, specifically Indonesian is a less common language to be tested for such studies. This study analyzes the comparative analysis of ChatGPT-4o and DeepSeek-R1 in processing multilingual translation and summarization NLP tasks using Indonesian and English data input. The study implements a mixed-approach method involving quantitative approach using BLEU, ROUGE, Perplexity, and response time, along with qualitative methods using Likert scale and A/B testing through expert assessments. The results show ChatGPT-4o consistently outperforms DeepSeek-R1 when it comes to overall summarization and translation tasks with a strong bias towards English sources that confirmed the issue of data and resource disparity within modern large language models. The findings in this study emphasizes the importance of evaluating Large Language Models in multilingual contexts and offers a more thorough analysis into the strengths and limitations of each model and language model bias which contribute to NLP development for both low- and high-resource languages.