Tesla has quietly pushed weekly Optimus output from a few dozen units in the second quarter to several hundred per week by August, an Electrek investigation published 25 September 2026 says — but the humanoid still cannot generalize across new tasks, and its hands remain a manual, screw-by-screw bottleneck.
From dozens to hundreds per week
Electrek reports that Tesla is now building Optimus V3 at a rate of several hundred robots per week, up from a few dozen weekly in Q2 2026, with a year-end target of more than 1,000 per week and a long-run ambition of roughly 20,000 per week that Elon Musk has repeatedly floated on earnings calls. The build sits on the converted Model S/X line at Fremont after Tesla wound the sedan program down in May.
Hands are still the hard part
The Optimus hand, with more than 100 hand-assembled components per unit, is the single biggest source of variability on the line, and Tesla's tactile sensors are still failing in ways that show up as reliability drift rather than clean binary faults. Electrek's sources say it still takes several days of teleoperation and reinforcement learning for the robot to internalize even a basic new task — a gap that becomes glaring the moment Optimus is asked to work with an unfamiliar object or in an unfamiliar cell.
500,000 hours of training footage — and a doubling plan
Tesla has amassed more than 500,000 hours of teleoperated training video, split across five dedicated Optimus training hubs in Colorado, Arizona and Florida, and internal targets call for that dataset to roughly double by year-end. Musk himself has admitted that his earlier promise of "several thousand Optimus doing useful work" inside Tesla by late 2025 landed at zero — a candid reset that helps explain why the current messaging focuses on line-rate and hand yield rather than on autonomy demos.
Where it fits the humanoid race
The ramp lands the same week rivals are showing their own scale ambitions: Toyota's $6.4B, 400,000-robot 2028 plan, 1X's 50,000-NEO shipment target for 2027, and Agility Robotics' push beyond Digit into wheeled form factors. The takeaway from Electrek's reporting: Tesla can pour more Optimuses out of Fremont than any Western rival, but until the hand yield curve and a working general-purpose policy show up in the same quarter, the ramp is manufacturing progress, not product-market fit.
Reporting based on coverage from Electrek and prior Tesla Q2 2026 disclosures.
