Raspberry Pi Launches AI HAT+ 2 for Local Generative AI

On January 15, 2026, Raspberry Pi released the AI HAT+ 2, a $130 hardware add-on for the Raspberry Pi 5 designed to execute generative AI large-language models (LLMs) and machine vision tasks locally. The board features a Hailo-10H NPU providing 40 TOPS (INT4) of inferencing performance and 8GB of dedicated LPDDR4X RAM. This architecture offloads AI workloads from the main CPU, supporting models such as DeepSeek-R1-Distill, Llama 3.2, and Qwen 2.5, typically ranging from 1 to 7 billion parameters.

While the device enables cloud-free computing for battery-powered robotics and real-time object detection, early performance reviews are mixed. Some benchmarks indicate that the Raspberry Pi 5 CPU may outperform the Hailo-10H NPU in specific tasks. YouTuber Jeff Geerling noted that a standalone Raspberry Pi 5 with 16GB of RAM might be more flexible and faster than the HAT+ 2, citing the board's 3W power limitation compared to the Pi 5's 10W capacity.

Additionally, early reports indicate the hardware is hindered by software bugs and inaccurate model outputs. Despite these issues, the company maintains that the AI HAT+ 2 provides computer vision performance broadly equivalent to its predecessor for tasks like pose estimation and scene segmentation.