Tesla Preps Gen 3 Optimus, Claims AGI Lead

The Gen 3 Optimus hand is a significant leap in dexterity, jumping from 11 to 22 degrees of freedom (DOF), nearing the human hand's ~27 DOF. By relocating all actuators to the forearm and using a tendon-driven system, the design reduces weight and increases agility, a crucial step for manipulating delicate or complex objects in a factory setting. Musk's "atom-shaping" concept frames the AGI race as one of physical-world interaction, not just language processing. Tesla's strategy is to leverage its vast real-world driving data and custom FSD computing hardware to give Optimus a foundational understanding of physical space, a dataset software-only AI labs at Google or OpenAI lack. The competitive landscape is intensifying, with multiple firms pursuing different AI strategies. Figure AI, after initially partnering with OpenAI, pivoted to developing its own vertically integrated AI, stating that outsourcing AI is insufficient for embodied robotics. Meanwhile, Boston Dynamics is collaborating with Toyota to train its Atlas robot using Large Behavior Models (LBMs) for complex, long-horizon tasks. This vertical integration of hardware and AI is becoming a key philosophical divide in the industry. The argument is that true progress in robotics requires a tight feedback loop between the physical body and the AI model, something a "full stack" approach enables by controlling everything from the custom actuators to the neural net training. Tesla plans to convert the Model S/X production lines at its Fremont factory into a pilot line for Optimus. The initial goal is to deploy over 1,000 robots within its own manufacturing facilities during 2026 to gather data and refine tasks before scaling to a target of 1 million units per year.