Samsung Pushes Deeper into On-Device AI Ahead of Galaxy S26 Launch

Samsung aims to bring faster, smarter, and more private artificial intelligence (AI) directly onto its devices. The company today shared new insights into its efforts to run powerful AI models directly on smartphones and other products without relying on cloud servers. Hopefully, the firm will include optimized on-device AI features in the Galaxy S26 series, expected to launch in January or February 2026.

Samsung advances on-device AI with model compression and runtime engine

Even though premium smartphones come with powerful chipsets, RAM, and storage, there are always challenges when it comes to running advanced AI features locally. Since a large AI model performs billions of computations directly on a device, it can drain battery life, generate heat, and slow performance.

To address this, Samsung is focusing on model compression that reduces the size of AI models without greatly affecting their accuracy. Thanks to the quantization process, it simplifies complex calculations into more efficient integer formats. This is similar to compressing a high-resolution photo, where the file size decreases while maintaining almost the same quality.

Samsung says that every device has different memory and processing power, so a general approach can’t deliver cloud-level AI performance. As a result, the company is creating its own compression algorithms to enable AI to run faster on each device. Furthermore, it is developing an AI runtime engine that manages how models run on a device’s processor. The runtime helps assign each task to the optimal chip while minimizing memory use.

Samsung is also exploring new AI model architectures for on-device environments. Most LLMs today rely on transformer architecture, which analyzes an entire sentence at once. However, the company is adopting a different approach, testing each one to see how efficiently it can work in real device environments.

“AI will become better at learning in real time on the device and adapting to each user’s environment,” said Dr. MyungJoo Ham, Master at AI Center, Samsung Research. “The future lies in delivering natural, individualized services while safeguarding data privacy.”