Every humanoid robot company covered in this series shows off polished demo footage, but the actual autonomy level behind those videos varies enormously once you look at verified deployment data rather than marketing reels — 1X is the most transparent about relying on human teleoperation for roughly a third of tasks, Figure has the most independently verified real-world working hours, and Tesla Optimus has repeatedly shown a wide, well-documented gap between what its demos imply and what its own CEO has publicly confirmed. None of this means the technology isn't progressing. It means the honest starting point for evaluating any of these companies is verified deployment data, not the highlight reel.
1X NEO: The Most Honest About Its Limits
1X, backed by OpenAI and EQT Ventures at a valuation near $10 billion, is the only humanoid robot company with a genuine "click here, enter your address" consumer purchase flow in 2026, priced around $20,000 or roughly $500 a month as a subscription. Industry estimates put NEO's launch autonomy at 60 to 70% — meaning roughly a third of tasks it encounters still require human intervention. For those cases, NEO uses what 1X calls Expert Mode: a human teleoperator at the company takes over remotely, completes the task, and the robot's model learns from that demonstration. 1X has been notably candid about this reliance on remote human backup while its AI continues learning, a level of transparency industry analysts have specifically credited as refreshing in a sector known for staged videos — even though it also means NEO ranks behind more autonomous-seeming rivals in some direct comparisons, specifically because of that teleoperation dependency.
Figure: The Most Verified Real-World Deployment
Figure AI has built the strongest evidence base of actual production deployment among its peers. Figure 02 helped build more than 30,000 BMW X3 vehicles over a 10-month deployment at BMW's Plant Spartanburg, logging more than 1,250 operational hours on a real production line for a paying customer — figures that come from BMW's own published numbers, not Figure's marketing. Figure is now expanding that relationship to BMW's Plant Leipzig with its newer Figure 03 platform. One comparison tracker scoring real-world traction specifically (not raw robot specs) gave Figure a score of 79 versus Tesla's 32, a 47-point gap the analysis attributed directly to Figure having a named paying customer already generating verified productive output, while Tesla had zero announced external customers as of April 2026.
Tesla Optimus: The Widest Demo-to-Reality Gap
Tesla's pattern is the most publicly documented example of the gap this article is about. Several high-profile demos have been confirmed teleoperated after the fact: the October 2024 "We, Robot" event's bartending and crowd-interaction segments were later confirmed to use human teleoperation, and Tesla's December 2025 Autonomy Visualized event in Miami saw an Optimus unit tip over with hand movements that closely resembled removing a VR headset — an incident flagged publicly by observers including Gene Munster and Robert Scoble as difficult to explain under a genuinely autonomous-operation scenario. Not every demo fits this pattern: Tesla's October 2025 kung fu demonstration was specifically confirmed as AI-driven rather than remotely piloted, a genuine autonomy proof point worth crediting. Inside Tesla's own factories, Optimus performs three confirmed autonomous tasks — battery cell sorting, parts handling, and quality inspection — all narrow, repetitive, and confined to structured environments. On Tesla's January 28, 2026 earnings call, Musk himself stated plainly that Optimus units are "not doing useful work" and remain in the "R&D phase," directly contradicting earlier, more bullish public framing. Program lead Milan Kovac left Tesla in June 2025, reportedly amid a pause tied to hand and arm reliability issues.
What "Deployable Autonomy" Actually Requires
It's worth being specific about the bar none of these companies have publicly cleared yet. Genuinely deployable factory autonomy requires eight-plus hours of unsupervised operation, robust error recovery when something goes wrong, real adaptation to unexpected situations rather than pre-scripted responses, production-speed task execution, formal safety certification for working alongside humans, reliable performance sustained across multiple shifts, and integration with charging and maintenance cycles. As of 2026, none of these specific capabilities have been publicly and independently demonstrated by any humanoid robot company, including the ones covered here — a useful checklist for evaluating the next round of demo videos rather than taking any single polished clip at face value.
Autonomy vs Deployment Evidence at a Glance
| Company | Estimated real autonomy | Verified deployment evidence | Transparency about limits |
|---|---|---|---|
| 1X NEO | ~60-70% at launch; Expert Mode teleoperation for the rest | First humanoid with a genuine direct-to-consumer purchase flow | High — company openly discusses teleoperation reliance |
| Figure | Not disclosed as a single percentage | 1,250+ verified hours, 30,000+ vehicles built at BMW Plant Spartanburg | Moderate — verified by customer's own published numbers |
| Tesla Optimus | Not disclosed; several demos later confirmed teleoperated | 3 confirmed narrow factory tasks; zero external paying customers (April 2026) | Low — inconsistent disclosure; CEO later contradicted earlier framing |
Frequently Asked Questions
How autonomous is 1X's NEO robot really?
Industry estimates put NEO's launch autonomy at 60 to 70%, meaning roughly a third of tasks require a human teleoperator through the company's Expert Mode. 1X has been notably transparent about this compared to some competitors.
Has Figure's BMW deployment actually been verified?
Yes. Figure 02 logged more than 1,250 operational hours helping build over 30,000 BMW X3 vehicles at Plant Spartanburg over a 10-month deployment, figures drawn from BMW's own published numbers rather than Figure's marketing.
Have Tesla Optimus demos actually used hidden teleoperation?
Yes, on multiple confirmed occasions. The October 2024 "We, Robot" event's bartending segments were later confirmed teleoperated, and a December 2025 fall during a Miami event showed hand movements resembling VR-headset removal, though Tesla's October 2025 kung fu demo was confirmed genuinely AI-driven.
What does full autonomous deployment actually require that no company has shown yet?
Eight-plus hours of unsupervised operation, robust error recovery, adaptation to unexpected situations, production-speed execution, formal safety certification, reliable multi-shift performance, and integration with charging and maintenance cycles — none of which any humanoid robot company had publicly and independently demonstrated as of 2026.