For the last two years, large language models have dominated the AI scene. But that might be ... [+] changing soon. (Photo by Kirill KUDRYAVTSEV / AFP) (Photo by KIRILL KUDRYAVTSEV/AFP via Getty Images)AFP via Getty ImagesWhen ChatGPT, Gemini and its other generative AI cohorts burst onto the scene a little over two years ago, talk about large language models artificial intelligence models trained on large volumes of datasets to understand and generate human-like texts and visuals dominated the technology scene. For years, the AI race was defined by scale bigger models, more data and greater compute.But lately, theres been a growing move away from large language models like GPT-4 and Gemini toward something smaller, leaner and perhaps even more powerful in certain business applications.The next wave of AI is being built for specificity,Jahan Ali, founder and CEO of MobileLive, told me in an interview. Small language models allow us to train AI on domain-specific knowledge, making them far more effective for real-world business needs.The Rise Of Small Language ModelsSLMs are AI models fine-tuned for specific industries, tasks and operational workflows. Unlike LLMs, which process vast amounts of general knowledge, SLMs are built with precision and efficiency in mind. This means they require less computation power, cost significantly less to run and, crucially, deliver more business-relevant insights than their larger counterparts.SLMs are not just scaled-down versions of LLMs, explained Ali. They are optimized to excel in specific domains whether its finance, healthcare, or software development. This allows them to deliver more accurate, reliable results tailored to the unique needs of an organization.Avi Baum, CTO and co-founder of Hailo, expanded on this idea and told me that When LLMs first emerged, they were designed to demonstrate intelligence at an unprecedented scale. But when practicality came into play, smaller, distilled models began to show up. These SLMs maintain strong reasoning capabilities while being efficient enough to run locally without reliance on cloud computing.Another reason why were now seeing a greater demand for SLMs, according to Baum, is that there are several privacy and security concerns linked to LLMs. Many enterprises hesitate to use cloud-based generative AI tools because of concerns about data leakage and compliance risks. With SLMs, businesses can deploy AI directly on edge devices, such as laptops, robots and mobile phones, ensuring their proprietary data remains protected.Small Language Models And Agentic AIThe conversation around small language models inevitably leans into the broader discussion on agentic AI a new wave of so-called AI agents that, unlike traditional AI systems, operate autonomously, making real-time decisions based on incoming data. To achieve such incredible feats, these agents need models that are lightweight, fast and highly specialized precisely where SLMs shine the most.As Stu Robarts noted in an article for Verdict, SLMs can be better suited to agentic AI due to the greater levels of accuracy that can be achieved with them compared to LLMs, greater operational efficiency through requiring less computing power and greater propensity for integration across ecosystems due to their smaller size and resource demands.Ali sees this as the next major breakthrough in AI. SLMs enable AI agents to make decisions with greater autonomy because they are trained on deep, domain-specific knowledge. Imagine a financial AI agent that doesnt just generate market insights, but actively executes trades based on real-time data. Or a logistics AI that not only tracks supply chains but autonomously optimizes delivery routes and inventory levels, he said.Shahid Ahmed, global EVP at NTT New Ventures and Innovation, also shares a similar vision. SLMs align with the broader trend of Agentic AI by allowing autonomous decision-making at the edge. In a smart factory, for example, an AI agent could use an SLM to proactively detect equipment failures, adjust machine settings, or schedule maintenance all without human intervention.This has massive implications across industries: From healthcare where SLMs can assist in diagnosing patients with greater specificity to customer service where they can power AI agents that truly understand industry jargon the applications are endless.The Business Case For SLMsOpenAI, Google and Anthropic have all poured billions into training their frontier large language models. While these models have been very helpful, being the foundational models from which researchers have distilled smaller models, many people believe the costs simply dont make sense and question the ROI on such massive dollar investments.Thats why the economics of AI development now seems to be shifting in favor of SLMs. According to Ahmed, the biggest advantage of SLMs is their cost-effectiveness.Large models require extensive computing power, which translates to higher operational costs. SLMs, on the other hand, consume fewer resources while delivering high accuracy for specific tasks. This results in a much higher return on investment for businesses, he said a point which Ali vehemently echoed, pointing out that the gap in ROI between LLMs and SLMs is becoming more apparent.Why pay millions to train and run a massive LLM when you can achieve better business outcomes with a smaller, cheaper model tailored to your exact needs?, Ali questioned.Challenges And Adoption StrategiesOf course, small language models arent without their challenges, especially when it comes to training them, which often requires high-quality domain-specific data. SLMs also sometimes struggle with long-form reasoning tasks that require broader contextual knowledge.Yuval Illuz, cybersecurity and AI expert and COO of OurCrowd, highlighted this data challenge in an interview with me: The key to making SLMs work is curating the right training data. Without high-quality datasets, an SLM can quickly become unreliable. The best approach is to continuously retrain models using real-world business data.However, in spite of these hurdles, Illuz believes that SLMs will be central to the future of enterprise AI. We are moving toward a hybrid AI world, where businesses will leverage both LLMs and SLMs in tandem. LLMs will remain useful for general knowledge, while SLMs will handle business-critical operations that demand accuracy, security and speed.The Quest For More ValueThe AI revolution started with the belief that bigger models meant better results. But now, companies are fast realizing that business impact is more important than model size. For many business leaders, the question isnt about which AI model people are jumping on, but about which model drives real business value for our company?As Ali noted, the future isnt just about building smarter AI its about building AI that actually works for businesses. And SLMs are proving that sometimes, less is more.