Raspberry Pi has officially shattered the “vision-only” barrier for edge computing. On January 15, 2026, the company announced the AI HAT+ 2, a powerful PCIe add-on for the Raspberry Pi 5. Unlike previous iterations focused on object detection, the new board is a dedicated “AI co-processor” capable of running sophisticated generative AI models entirely offline.
The 8GB RAM Breakthrough
The most striking feature of the AI HAT+ 2 is the inclusion of 8GB of LPDDR4X-4267 RAM directly on the add-on board.
Historically, AI accelerators on the Pi were limited by the host’s memory bandwidth. By putting 8GB of dedicated memory on the HAT itself, Raspberry Pi has ensured that:
- Large Models Fit: Generative models like Llama 3.2-3B and Qwen2.5-Coder can sit entirely within the accelerator’s memory.
- Zero Host Drag: The Raspberry Pi 5’s system RAM remains completely free to handle the OS, desktop, and other applications.
- Low Latency: Data doesn’t need to “hop” back and forth across the PCIe link as frequently, significantly boosting token-per-second performance.
Technical Specifications: Powered by Hailo-10H
The board is built around the Hailo-10H neural network accelerator, a massive upgrade over the Hailo-8 found in the original AI Kit.
| Feature | AI HAT+ (2024) | AI HAT+ 2 (Jan 2026) |
| Processor | Hailo-8 / 8L | Hailo-10H |
| Performance | 13–26 TOPS | 40 TOPS (INT4) |
| Onboard RAM | None | 8GB LPDDR4X |
| Primary Focus | Computer Vision (YOLO) | GenAI, LLMs, & VLMs |
| Price | $70 – $110 | $130 |
What Can You Run?
At launch, the AI HAT+ 2 supports a curated selection of “edge-ready” models that can be deployed with minimal configuration:
- Text-to-Text: Llama-3.2-3B-Instruct and Qwen2.5-Instruct.
- Vision-Language (VLM): Qwen2.5-VL-3B-Instruct (allows the Pi to “see” and describe scenes in natural language).
- Speech-to-Text: High-speed, local Whisper transcription.
Conclusion: A Professional Tool for $130
Priced at $130, the AI HAT+ 2 is positioned as a professional development tool rather than a casual hobbyist toy. It is supplied with an optional heatsink (highly recommended for LLM workloads) and is compatible with the official Raspberry Pi Active Cooler. For developers looking to build secure, offline AI agents or advanced robotics, the AI HAT+ 2 transforms the Raspberry Pi 5 from a simple computer into a high-performance AI workstation.
