2026-07-19
Written by Lena Kaplan
Small AI models are revolutionizing industries with their efficiency and cost-effectiveness. From healthcare to finance, these compact AI systems are being integrated into various sectors to drive innovation and improvement.
In recent years, small artificial intelligence (AI) models have been gaining traction around the world. These miniature marvels are being used to tackle pressing problems in healthcare, agriculture, and other fields, often without the need for expensive infrastructure or internet connectivity.
Adebayo Alonge's RxScanner is a handheld spectrometer that scans pills with infrared light and uses an AI model to identify their molecular profile or report if they're counterfeit. In 2019, Alonge was preparing to demonstrate his device in a Cape Town hotel room, but the system failed due to limited bandwidth. Undeterred, he asked his engineers to shrink the AI model down to fit on his Android phone, and they delivered.
This experience sparked Alonge's interest in small AI, which is defined as language models with at most a few billion parameters. These models can run directly on devices like smartphones or Raspberry Pis, making them ideal for deployment in areas without access to cutting-edge technology.
Advocates like Marcelo José Rovai argue that small AI is the future of AI development, particularly in the Global South. "This is the most important area in AI nowadays," says Rovai, a professor at the Institute of Engineering and Information Systems at the Federal University of Itajubá. "It's growing very fast."
One reason for this rapid growth is the proliferation of low-power devices that can run small AI models. In 2025, more than a third of all smartphones shipped worldwide were capable of running generative AI, and this figure is expected to reach 45 percent by the end of this year.
Another factor contributing to the rise of small AI is the shrinking footprint of language models. Devices like Google DeepMind's Gemma 4 and Alibaba's Qwen 3.5 are "fantastic" for small AI, allowing users to adjust connections between parameters to suit their needs.
The World Bank now actively promotes small AI development with grants, mentorship programs, financing, technical advice, and government policies that support this technology. In Rwanda, the World Bank is backing a government program to help low-income households get devices that can run AI.
However, experts caution that large language models are not going away entirely. To create generative AI that can run on small devices, we need the architectural insights, data processing, and results of larger models.
Moreover, implementing small AI won't automatically solve the challenges of development and digital inequality. Reliable power, a supply chain that works, and an educational system that develops the talents needed to create AI tools are still essential for widespread adoption.
Despite these caveats, small AI has the potential to bring technology to millions of people worldwide, particularly in areas without access to cutting-edge infrastructure. As Alonge says, "This is not just a promising trend; it may be the form of AI that will touch the most lives and remain sustainable in the long term."
As small AI continues to grow, we can expect to see more innovative applications across various industries. From detecting diseases in cashew plants to identifying ant infestations in vineyards, these miniature models are being used to tackle pressing problems without the need for expensive infrastructure.
In parts of the world where the future of small AI is uncertain, experts urge policymakers to invest in infrastructure that will support this technology long-term. As Alonge says, "The question is whether or not the political actors are wise enough to invest in infrastructure to support it."
For now, small AI remains a beacon of hope for those seeking to harness the power of technology to drive positive change. With its potential to bring tech to the Global South and tackle pressing problems without expensive infrastructure, this emerging field is sure to continue its rapid growth and have a lasting impact on our world.
Small AI's potential to benefit people in areas with limited access to big AI cannot be overstated. For example, A drone-based system developed by Bala Murugan and colleagues at the Vellore Institute of Technology takes photos of cashew plants and quickly identifies those with splotches that indicate disease. All the processing takes place on the drone itself, so there’s no need for a computer on-site, nor for a connection to a central server.
In Uruguay, an ant infestation system developed by researchers has been used to detect pests in vineyards, saving farmers from financial losses. In Brazil, small AI models are being used to analyze electrocardiograms from devices that can't run more complex equipment, improving healthcare outcomes for millions of people.
These examples demonstrate the potential of small AI to deliver life-saving services and improve daily lives, often without the need for expensive infrastructure or internet connectivity.
The benefits of small AI extend beyond the Global South. In India, where the government's AI plans call for the development of small AI, many such systems are working for farmers. These systems can help farmers identify diseases in their crops more efficiently, reducing waste and improving yields.
In addition to these applications, small AI models have been used to analyze water quality, detect breast cancer through mobile devices, and even track air pollution levels in real-time.
The key advantage of small AI is its ability to run on a variety of low-power devices, making it accessible to people who may not have access to more powerful hardware. In some cases, small language models are trained from scratch for specific tasks, allowing them to be highly specialized and efficient.
Another benefit of small AI is its relatively low cost. Unlike large language models, which require significant resources and infrastructure, small AI models can run on devices that are affordable and widely available.
Despite these advantages, small AI still has limitations. For example, some applications may not be suitable for small AI, requiring more powerful hardware or specialized software.
Moreover, small AI models may not always be able to replicate the accuracy of large language models. While small AI can deliver highly accurate results in specific contexts, it may not be as effective in more complex or generalizable tasks.
In summary, small AI has the potential to bring technology to millions of people worldwide, particularly in areas without access to cutting-edge infrastructure. With its ability to run on low-power devices and deliver life-saving services, small AI is an exciting development that could have a lasting impact on our world.
