Small language models are gaining traction in corporate AI strategies

A survey by consulting firm IDC, cited by Canaltech, predicts that by 2026, nine out of ten language model use cases among the top 1,000 companies in the Asia-Pacific region will be based on Small Language Models (SLMs), or small language models.

What are small models?

SLMs have fewer parameters and typically require less computing power than the large models used by services such as ChatGPT and other generative assistants.

This difference doesn’t mean that a small model is simply a worse version of a large model. The goal is to choose an artificial intelligence that is tailored to the specific task.

An organization may use a larger model for tasks that require complex reasoning and a smaller model for more repetitive and specific functions.

This specialization can be crucial when a company needs to process large volumes of requests. The lower the computational demand of each operation, the lower the execution cost tends to be.

Cost and speed are also factors

Implementing AI on a large scale involves more than just paying for access to a chatbot. Each request consumes processing resources, and a corporate application can perform thousands or millions of them.

Smaller models can reduce this demand and also respond more quickly in certain tasks.

A recent analysis of studies on SLMs identified gains specifically in computational efficiency and in the possibility of deployment closer to the company’s own infrastructure.

The choice can also be financial. If a simple task can be solved by a smaller AI, using a large-scale model for all requests can represent an unnecessary expense.

Data and privacy also play a role

Another reason for companies to consider SLMs is control over data.

Smaller models can be installed on private servers or in environments controlled by the organization itself, depending on the chosen architecture.

This is important for companies that handle financial information, internal documents, customer data, or other content that should not circulate freely in external services.

However, SLMs are not automatically more secure. Protection depends on how the model is configured, updated, monitored, and connected to the company’s data.

The trend is to combine different models

The growth of small models does not mean that giants will disappear. An architecture can distribute tasks among different artificial intelligences.

A simple request can be forwarded to a smaller model, while a task that requires greater reasoning capacity is directed to a more robust model. The result is a kind of division of labor among AIs.

For companies, the discussion is no longer just about which model is the smartest, but includes a more practical question: which model can perform a given task with sufficient quality, at the lowest cost, and with the necessary level of control?

It is this equation that helps explain why smaller models are gaining ground. In the corporate race for artificial intelligence, size is no longer the only criterion for deciding which technology makes sense to adopt.