LLM-Driven Embodied AI with the Reachy Mini

This is the Reachy Mini: the first open-source desktop robot for embodied AI and Human-Robot Interaction (HRI). Developed by Pollen Robotics and Hugging Face, this 11-inch, 3.3 lb unit is a powerful, accessible platform starting at $299. It features a 6 DoF head, full body rotation, and a comprehensive Python SDK for immediate coding. Developers integrate seamlessly with the Hugging Face Hub, accessing over 1.7 million AI models for vision and speech tasks. Choose the Lite (wired) or Wireless (onboard Raspberry Pi 5) model: both deliver a robust environment for prototyping and sharing real-world AI applications directly from your desk.

The Reachy Mini Python SDK lets you program and control Pollen Robotics' compact, expressive robot using clean, direct Python code. Compatible with Python 3.10 and higher, the SDK manages hardware communication over USB or Wi-Fi, abstracting complex motor coordination into straightforward commands like goto_target. You can easily direct head movements, rotate antennas, capture camera frames as NumPy arrays, and access IMU data. Additionally, the SDK integrates smoothly with Hugging Face Spaces and modern AI coding agents, allowing you to rapidly deploy large language models and computer vision pipelines directly onto your physical or simulated robot.

Qwen3 is a powerful, open-weighted LLM (Apache 2.0) from Alibaba Cloud, engineered for peak performance and efficiency. Its core design features a diverse model lineup, including dense models (0.6B to 32B) and efficient MoE variants like the Qwen3-235B (22B active parameters). The key innovation is the Hybrid Thinking Mode, which dynamically toggles between deep, step-by-step reasoning and fast, non-thinking responses. This model supports an extensive 119 languages and handles long-context tasks up to 128K tokens, making it a robust, versatile choice for advanced multilingual and agentic workflows.

Llama 3.1 delivers state-of-the-art performance through a flagship 405B parameter model trained on 15 trillion tokens. It supports a 128k context window: ideal for analyzing massive datasets or long-form documentation. Developers utilize Llama for diverse tasks (multilingual translation, Python code generation, and complex reasoning) while maintaining data sovereignty via local hosting. The ecosystem includes the Llama Stack for agentic workflows and optimized weights for 8B and 70B models, ensuring high throughput on consumer hardware or enterprise clusters.

The Hugging Face Inference API bypasses the headaches of local hardware configuration and model deployment by offering immediate, serverless access to a massive catalog of open-source AI models. Developers can integrate tasks like text generation, image classification, and embeddings into their applications using simple HTTP requests or official Python and JavaScript clients. The API handles the underlying infrastructure automatically, scaling seamlessly from quick prototyping to high-throughput production environments while supporting top-tier open weights (including Llama, Mistral, and Stable Diffusion) with minimal setup.