Abstract

The development of Generative Artificial Intelligence is revolutionizing human–machine interaction models. In this context, large language models (LLMs) have emerged as tools capable of learning complex patterns. However, many of these technologies are highly resource intensive. An alternative to these complex models is the advent of Small Language Models (SLMs). These smaller models process fewer parameters but achieve acceptable performance in their responses, striking a balance between cost and quality. This study characterizes different SLMs to facilitate decision-making in their implementation. In the methods section, a systematic review is conducted, serving as a guide for researchers and professionals seeking to select the most suitable SLM for their specific needs. An analysis of the efficiency of these models contributes to the application of Artificial Intelligence techniques from a sustainability perspective. The results section presents a comparison of various SLMs available on the Ollama platform. The models compared include Qwen2.5, Phi3.5, Mistral-small, Llama3.1, and Gemma2. A comparative analysis evaluates these models based on their efficiency and effectiveness in terms of computational resources and the human effort required to develop task-specific conversational systems. The study demonstrates the feasibility of using these smaller models in various decision-making environments. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.