Raspberry Pi has unveiled the Raspberry Pi AI HAT+ 2, an add-on board for Raspberry Pi 5 that adds support for local generative AI workloads, including large language and vision-language models, for offline and low-latency edge applications.

The AI HAT+ 2 is based on the Hailo-10H neural network accelerator, delivering up to 40 TOPS of INT4 inferencing performance. While designed primarily for generative AI workloads, computer vision performance remains comparable to the earlier 26 TOPS AI HAT+, supporting tasks such as object detection, pose estimation, and scene segmentation.

Raspberry Pi AI HAT+ 2 Installation

A key hardware change compared to the original model is the addition of 8 GB of dedicated on-board RAM. This allows the AI HAT+ 2 to handle significantly larger models than before, including compact LLMs and vision-language models (VLMs), while leaving the host Raspberry Pi 5 free to manage system-level tasks.

Raspberry Pi AI HAT+ 2 Dimensions

At launch, supported models include DeepSeek-R1-Distill (1.5B parameters), Llama 3.2 (1B), Qwen2, Qwen2.5-Instruct, and Qwen2.5-Coder variants, with additional and larger models expected in future updates. These models are designed to run entirely on-device, avoiding reliance on cloud services for inference.

Like its predecessor, the AI HAT+ 2 connects to the Raspberry Pi 5 using the PCI Express interface and conforms to the Raspberry Pi HAT+ specification. When used with an up-to-date Raspberry Pi OS image, the accelerator is automatically detected and integrated into the system.

The board also maintains tight integration with Raspberry Pi’s camera software stack, including libcamera, rpicam-apps, and Picamera2, enabling accelerated post-processing for vision-based pipelines.

AI HAT+ 2 Heatsink Dimensions

For developers looking to customize workloads, the AI HAT+ 2 supports retraining of vision models and Low-Rank Adaptation (LoRA)–based fine-tuning of language models using the Hailo Dataflow Compiler.

This allows task-specific adaptation of pre-trained models while keeping most parameters fixed, enabling applications such as speech-to-text, translation, and scene analysis within the constraints of edge hardware.

AI HAT+ 2 + Raspberry Pi 

The board is supplied with an optional heatsink, which is recommended for sustained or intensive workloads to prevent thermal throttling. It is designed to operate in ambient temperatures from 0 °C to 50 °C and is compatible with Raspberry Pi 5 systems fitted with the official Active Cooler.

Further Information

The Raspberry Pi AI HAT+ 2 is available now with a list price of $130, and Raspberry Pi has indicated a production lifetime extending through at least January 2036.

The board is already available from distributors including The Pi Hut (£124.80), Pimoroni ($138.32), and PiShop.us ($130.00).