The Raspberry PiAI HAT+ 2 adds local generative AI capability to Raspberry Pi 5, making it possible to run large language models (LLMs) and vision-language models (VLMs)directly on the board. Built around the Hailo-10H AI accelerator and equipped with 8GB of dedicated on-board RAM, it allows demanding AI workloads to run locally while leaving the host Raspberry Pi 5 free for other processing.
The Hailo-10H accelerator delivers 40 TOPS of INT4 inferencing performance, with computer vision performance comparable to the original 26 TOPS Raspberry Pi AI HAT+. This level of performance is suited to running generative AI and vision-based models efficiently, using the sample models provided by Hailo, or create your own custom-trained models for tasks such as speech-to-text, translation, and visual scene analysis.
The Raspberry Pi AI HAT+ 2 connects via the Raspberry Pi 5’s PCI Express interface and is automatically detected when running an up-to-date Raspberry Pi OS image. It integrates directly with Raspberry Pi’s camera software stack, allowing supported rpicam-apps post-processing tasks to make use of the accelerator without additional configuration.
The board conforms to the Raspberry Pi HAT+ specification and is supplied with an optional* HAT heatsink, plus a 16mm stacking header, spacers, and screws for fitting on a Raspberry Pi 5 with the Raspberry Pi Active Cooler installed.
*Whilst the heatsink supplied is optional, users running benchmarks or very intensive AI workloads are advised to install it to prevent thermal throttling.
Features
Hailo-10H AI accelerator delivering 40 TOPS (INT4) inferencing performance
Runs generative AI models efficiently using 8GB on-board RAM
Performance for computer vision models comparable to the Raspberry Pi AI HAT+ (26 TOPS)
Fully integrated into Raspberry Pi’s camera software stack
Conforms to Raspberry Pi HAT+ specification
Compatible with Raspberry Pi 5 only (requires PCIe interface)
Skip to content
