Downloads: 0
Comparative Study | Computer Science and Engineering | Volume 15 Issue 7, July 2026 | Pages: 2355 - 2360 | India
Comparative Study of Open-Source Large Language Models
Abstract: Large Language Models (LLMs) are now a significant advance in the field of artificial intelligence, demonstrating a solid capacity to comprehend, create, and think about human language in a natural and intuitive manner. The work gives a comparison of the open-source LLMs and popular models, including LLaMA and Mistral among other recent systems. It examines such pertinent issues as model design, model size, mode of training, transparency and benchmark performance. Nevertheless, recent studies have indicated that proprietary models are more likely to perform better, although open-source LLMs are more versatile, are more available and reproducible. In addition, the architectural constructions like Mixture-of-Experts (MoE), and effective attention mechanisms have also contributed to the capacity of the models and reduced the level of computation. However, standardized evaluation, trade-offs among hardware and efficiency, and deployment constraints are some of the few challenges that should be overcome. These problems will be addressed by the author in this paper, who is going to propose a system of comparisons, which will be structured in the context of performance, efficiency, and usability. The findings enable the identification of the suitable models to implement them into the practice and indicate the gaps in the research to develop more effective, clear, and scalable language models.
Keywords: Large Language Models, Open-Source LLMs, Transformer Architecture, Model Comparison, LLaMA, Mistral, Mixture-of-Experts, Natural Language Processing, Model Efficiency, Benchmark Evaluation, Fine-Tuning, Model Transparency, Computational Performance
How to Cite?: Pinak Tiwari, "Comparative Study of Open-Source Large Language Models", Volume 15 Issue 7, July 2026, International Journal of Science and Research (IJSR), Pages: 2355-2360, https://www.ijsr.net/getabstract.php?paperid=SR26726161252, DOI: https://dx.doi.org/10.21275/SR26726161252