Small is becoming very large in the world of AI.
Small language versions ( SLMs) are increasingly being used by many software companies, especially those looking to quickly ramp up their AI options. SLMs require less computing power and memory, leading to smaller data. These models, which were created for specific business tasks, are not only quicker to teach and build, but they are already outperforming or matching similarly sized models, which is good for any company that wants to apply AI, especially those with limited resources, funds, or period constraints. Over the next five years, the marketplace for SLMs is anticipated to grow at a constant 15 %.
On the flip side, the more well-known large language models ( LLMs) used in many AI applications are trained with massive datasets. This data may take months to teach, and it’s just the start — it is often followed by individual fine-tuning. LLMs can get significant financial burdens for most software companies and businesses because they involve substantial development costs that could exceed many million dollars, according to some estimations.
Since SLMs are growing in popularity, what’s future?
SLMs are the preferred choice for many companies because they use significantly fewer parameters and can be created from scratch or adapted from LLMs. They can be valuable to businesses looking for qualified quick wins. These models can become hosted in an enterprise data centre rather than the sky because they are smaller. SLMs are even more effective when open-source, and by teaching on carefully curated business datasets, they can be filtered for disagreeable articles with important concerns like management, risk, privacy, and bias mitigation, as this becomes extremely important in 2025 and above.
When it comes to AI, schedule is all
SLMs find a nice place among the numerous use cases when predicting outcomes from time series data. Timing is important in business, where every organization has a forecast of sales, demand, revenue, and capacity requirements, this is called time series forecasting, and it involves predicting potential values based on past observations collected in regular time intervals, whether that is normal, monthly, quarterly, or annually.
With a faster foundation model for this kind of multivariable forecasting, AI is anticipated to speed up and tighten business planning. For instance, an SLM called Tiny Time Mixers ( TTMs) can swiftly generate time-dependent outputs, predicting future trends in diverse domains such as electricity consumption, traffic congestion, retail, and finance. A global leader in the field of AI-powered investment solutions, QuantumStreeet AI, is using this kind of model to help its platform forecast stock price movements across industries by utilizing ESG data and sentiment signals from news and other data sources.
As innovation expands, models will be trained on even more data, produce stronger performances, and offer more flexibility with support for rolling forecasts and external variables.
Embedding AI into your daily operations today
Business is beginning to change as a result of AI. However, the breathless hype about AI of the past two years must be leavened with cost, trust, and resource considerations.
In fact, businesses may soon favor a combination of LLMs and SLMs, using bigger models first to solve some of the most difficult business issues before switching to smaller models that can provide the same results at lower costs and with lower latency.
Looking forward, SLMs will also play a prominent role in the advancement of AI agents that are capable of greater autonomy, sophisticated reasoning, and multi-step problem solving. SLMs feature support for key agentic capabilities, such as advanced reasoning and specific function calling, which are critical to ensure an agent can connect with external APIs, reassess its plan of action, and self-correct.
Enterprises that use artificial intelligence must strike the ideal balance between powerful and practical. Consider using an SLM as a race car and a LLM as a motorhome, both of which will take you where you want to go but have different purposes. The models that offer the best performance in terms of performance relative to model size while maximizing safety, speed, and cost-efficiency can be easily integrated into a variety of business settings and workflows.
SLMs will have a significant impact on your company’s ability to quickly implement AI across your business, whether your company is currently testing AI projects or considering using AI agents for business purposes tomorrow.
Raj Datta is Vice President, Software and AI Partnerships at IBM, where he spearheads strategy, sales, and strategic alliances. Before, he co-founded and was CEO at software company oak9, and was President of Software AG, North America. Prior, he spent 19 years at IBM in global and national leadership. Datt is a graduate of the University of Illinois, Urbana, with a BA in Economics and an MBA in Marketing and Finance from the Kellogg School of Management.
