Tesla has disclosed that Optimus Gen 3 units reached 1,000 deployed robots inside its Fremont facility by June 2026, with a target of 5,000 by year-end. Six months of production data is enough to assess where the humanoid robot is genuinely useful and where the limitations remain significant.
The tasks where Optimus Gen 3 is performing reliably: repetitive pick-and-place operations with standardised parts, trolley transport between work cells, and visual quality inspection on defined pass/fail criteria. These are high-repeatability, low-variance tasks where the robot's cycle time (roughly 85 percent of human pace on learned tasks) is acceptable and its failure rate (under 0.5 percent per cycle on trained motions) meets production quality standards.
The tasks where performance remains insufficient for unsupervised operation: cable harness routing, connector insertion requiring fine force feedback, and any operation that requires adapting to part variation outside trained tolerances. Tesla engineers have been candid in internal briefings that these tasks represent the hard frontier — not because the hardware cannot do them in principle, but because the training data volume required to achieve acceptable success rates in a production environment is still being accumulated.
The dexterity numbers: Optimus Gen 3 has 22 degrees of freedom in its hands, up from 11 in Gen 2. The hand redesign uses tactile sensors on fingertips that provide force feedback sufficient for delicate assembly — in a controlled lab setting. In the factory environment, with part tolerance variation and the added complexity of gloves or protective equipment requirements, the effective dexterity advantage is more modest than demo videos suggest.
Tesla's competitive position: no other company has humanoid robots in continuous production deployment at comparable scale. Figure AI has pilots at BMW's Spartanburg plant but at under 100 units. Boston Dynamics' Atlas Electric is in customer trials but primarily for materials handling, not assembly. Tesla's advantage is manufacturing scale and the data flywheel that comes with it — each robot generates telemetry that improves the next training cycle.
The realistic near-term picture: Optimus will handle an expanding set of defined, repetitive tasks inside Tesla's own facilities. Broad third-party deployment at scale is a 2027 to 2028 story, contingent on solving the variance adaptation problem.