IBM has introduced the Granite 4.0 Nano series of AI models, tailored explicitly for Internet of Things (IoT) edge computing and all-encompassing localized deployment across diverse scenarios. This series encompasses four models, with parameter counts spanning from 350 million to a staggering 1.5 billion. This marks a remarkable two-order-of-magnitude enhancement compared to rival offerings in the market. The core technologies underpinning these models encompass hybrid state space architectures, among other cutting-edge innovations.

The most compact model within the series is capable of running seamlessly within web browsers. Moreover, the 1-billion-parameter (1B) version surpasses the comparable Qwen3-1.7B model in terms of overall performance scores. These models boast enterprise-grade capabilities, including fully localized inference loops, and exhibit deep integration with a suite of toolchains. At present, IBM is actively engaged in the development of larger-scale versions of these models, as well as specialized models geared towards logical reasoning tasks.