The Berkeley Humanoid represents a major step toward democratizing humanoid robotics research by providing a fully open-source, low-cost, 3D-printed platform optimized for reinforcement learning. Standing at roughly one meter tall and costing under $10,000 in parts, the robot is designed to be assembled in academic labs without requiring heavy machining or expensive actuators. Its modular actuator system relies on brushless DC motors with custom planetary gearboxes, offering high power-to-weight ratios suitable for dynamic walking. The mechanical design emphasizes transparency and repairability: nearly every structural component can be printed on standard FDM printers, lowering the barrier for labs that cannot fabricate aluminum or carbon-fiber parts. The onboard compute stack pairs a Jetson Orin for perception and neural-network inference with a dedicated microcontroller for real-time motor control at multi-kilohertz bandwidths. Communications are mediated over a flexible EtherCAT-style bus that simplifies wiring and enables fast low-level closed-loop control. In terms of software, the Berkeley Humanoid ships with a full simulation stack in Isaac Gym (later Isaac Sim/MuJoCo) along with reference RL policies implemented via RSL-RL. This enables researchers to train locomotion behaviors in simulation and deploy them zero-shot on the physical robot. Locomotion studies demonstrate stable walking on flat terrain, mild slopes, and even limited stair traversal after minimal sim-to-real tuning. The broader impact of this platform is educational and collaborative: by lowering the cost floor for humanoid hardware, it invites more institutions to contribute to the growing body of work in legged locomotion, manipulation, and whole-body control. The documentation is extensive, and the community actively reports modifications, making the platform a living benchmark rather than a static hardware release.