Signals Inbox·July 15, 2026·Humanoid Robotics

Tesla Optimus Tracker - Last Update

Tesla Optimus is now a serious pre-production humanoid backed by a real factory plan, but Tesla still has not proved that the robot can work autonomously, reliably or profitably through a normal industrial shift.

Q1Where is Tesla Optimus today?

Tesla Optimus is currently an advanced pre-production robot with credible hardware, serious factory backing and almost no public proof of economic productivity.

We can now rule out the idea that Optimus is merely a publicity experiment. Tesla has developed several working generations, gathered robot-training data at scale, reserved major manufacturing space in Fremont and started preparing a production line designed around one million units of annual capacity. Tesla also says a much larger second-generation line is planned for Texas.

However, Tesla still gives outsiders remarkably little evidence about what Optimus accomplishes during a normal working day. We do not know how many useful cycles its robots complete, how often an operator intervenes, how long the hands last, how frequently robots fall or stop, or whether any current task saves Tesla money.

That lack of operating data has become more noticeable as competitors disclose harder results. Figure says two Figure 02 robots accumulated more than 1,250 operating hours at BMW, loaded over 90,000 parts and contributed to the production of more than 30,000 vehicles. Agility Robotics has already converted a logistics pilot into a multi-year commercial deployment with GXO. Tesla has announced a vastly larger production ambition without publishing an equivalent factory-performance record for Optimus.

The best description of Optimus today is production-intent, but not production-proven. The robot is far enough along for Tesla to industrialize it. We still cannot verify that the current machine deserves to be industrialized at the scale Tesla proposes.

Tesla Optimus status, July 2026

Q2Is Tesla Optimus doing real factory work today?

Tesla Optimus is working inside Tesla facilities today, but Tesla has still not shown that those robots are productive workers rather than development machines.

Tesla has repeatedly described its factories as the first deployment environment for Optimus. That choice makes sense. The company controls the building, the workflow, the employees and the tolerance for failure. Every attempted task can generate training data, even when the robot completes it too slowly to be economically useful.

Public demonstrations have shown Optimus sorting battery cells, moving objects, carrying trays and manipulating factory-related components. Tesla has also hired people specifically to collect physical training data and provide equipment feedback for the program. These roles point to an active loop between human demonstrations, robot trials and engineering changes.

What Tesla has not released is a result comparable with Figure’s BMW deployment. Figure disclosed runtime, parts handled, shifts worked, distance travelled and the number of vehicles touched by the production process. Tesla has disclosed none of those figures for Optimus, despite operating the robots in its own factories and facing fewer customer-confidentiality constraints.

The omission is becoming informative. Tesla is usually eager to advertise a manufacturing advantage when the numbers are strong. No throughput, no uptime, no intervention figures. That suggests the current deployments remain primarily useful to the Optimus engineering team.

Our verdict today is real factory testing without proven factory economics. The robots are probably doing genuine tasks, but the available evidence does not show that Tesla would keep them on the floor if they stopped generating development data.

Q3Can Tesla Optimus actually work by itself now?

Tesla Optimus can currently execute selected behaviors autonomously, but there is no evidence that it can complete a varied factory shift without regular human help.

Tesla’s approach uses cameras, neural networks and learned control policies rather than carefully programming every movement. The company presents Optimus as an extension of the vision-and-planning strategy developed for its vehicles.

The robot’s recent movements support part of that claim. Optimus can walk, recover its balance and perform learned manipulation sequences that would be difficult to reproduce through simple playback. Tesla’s continued investment in embodied data collection also shows that it is trying to convert human demonstrations into reusable behavior.

Still, we cannot calculate the number that counts: how many robot-hours require one human-operator hour. Tesla’s public interactions have included remotely assisted robots, and reporting on the internal program says the machines still often rely on teleoperation. Tesla has never published an intervention rate for factory deployments.

A robot that completes 95% of a task alone and needs a quick remote correction for the remaining 5% may already be useful. A robot that needs one dedicated operator watching continuously is basically an expensive remote-controlled machine.

Optimus appears to have narrow task autonomy, particularly for practiced movements in controlled spaces. Broad workplace autonomy remains unproven. Tesla has not shown the robot handling interruptions, misplaced parts, changing tools and several unfamiliar situations in a row over a full shift.

Q4Are Tesla Optimus hands finally good enough?

Tesla Optimus has a much more ambitious hand design now, but there is still no public proof that it can survive industrial work.

Tesla has correctly identified hands as one of the program’s hardest problems. Walking gets more attention online, yet useful manipulation decides whether Optimus can handle tools, route cables, pick irregular parts and perform jobs that fixed automation cannot easily absorb.

The latest production-intent generation is expected to use Tesla’s redesigned hand, with substantially more articulation than earlier versions. Tesla has also continued recruiting dedicated hand-design engineers. Clearly, the hand is still being treated as a specialized product of its own.

That recruiting activity tells us two things at once. There is a serious engineering effort, but hand development is still active close to the proposed manufacturing ramp. Late changes to fingers, actuators or tendon routing can ripple through suppliers, tooling, software and repair procedures.

The supply-chain interruption around rare-earth magnets exposed another weakness. Musk acknowledged that Chinese export restrictions had affected Optimus because powerful magnets are required in the robot’s actuators. A humanoid contains a lot of tightly packaged motion systems. One constrained component can hold up the whole machine.

Tesla has released no figures for grip accuracy, maximum fingertip force, actuator life, damaged-object rates or successful cycles before maintenance. Videos of delicate manipulation look good, but they tell us very little about factory readiness.

We currently rate the Optimus hand as technically promising and commercially unvalidated. Tesla may have solved enough dexterity for selected tasks. Durability, manufacturing yield and repair cost are still open.

Q5Does Tesla Optimus walk well enough for a factory?

Tesla Optimus walks well enough to be taken seriously today, although its public mobility record is still closer to a robotics demonstration than an industrial qualification test.

The improvement since Tesla’s first prototype is substantial. Optimus now walks without external supports, coordinates its upper and lower body and handles disturbances with much smoother balance corrections. Tesla has crossed the basic threshold where locomotion itself threatened to invalidate the project.

Factories demand something more boring. A useful robot has to cross changing surfaces while carrying parts, stop predictably near workers, survive minor collisions and repeat those movements thousands of times without frequent resets.

Tesla gives us no fall rate, distance between failures, safe walking speed around people or energy consumption under load. Figure, by comparison, reported that its BMW robots covered roughly 1.2 million steps and more than 200 miles during their deployment. It is still company-reported data, but at least it gives us an order of magnitude.

The competitive bar is rising too. Boston Dynamics has designed the production Atlas for Hyundai factories, while Hyundai is funding a US robotics facility with capacity for 30,000 robots per year. Atlas has not proved large-scale factory productivity either, but Boston Dynamics brings a much longer record in dynamic mobility, and Hyundai plans an automotive deployment by 2028.

Optimus mobility is credible enough for deployment experiments. Tesla’s remaining challenge is making walking an unremarkable, low-maintenance part of every shift.

Q6Is Tesla Optimus learning fast enough?

Tesla Optimus is probably collecting more embodied data than most competitors, but the current training process still appears heavily dependent on human labor.

Tesla has an unusual advantage here: it owns factories filled with tasks, equipment and potential training environments. It can place prototype robots near engineers, collect failures and change the workflow without negotiating with an external customer.

The company is also hiring operators whose job is to collect movement data, assist engineering operations and report equipment feedback. Earlier recruiting descriptions asked operators to wear motion-capture equipment and reproduce actions for the robot. That gives Tesla a direct stream of demonstrations, but it is expensive and slow compared with collecting internet-scale text or video.

The wider industry is already trying to reduce that dependence. In early 2026, 1X said it was shifting more training toward world models and robot-generated experience because human-operated demonstrations limit how quickly a dataset can grow. Its robots may still need remote assistance, but the strategic direction highlights the same bottleneck Tesla faces.

Tesla’s scale advantage will become meaningful when one learned task transfers across many robots, environments and object variations. Repeating the same human demonstration thousands of times could build a huge dataset without producing broad intelligence.

Today, we see a serious data engine with unknown learning efficiency. The hiring footprint tells us Tesla is collecting aggressively. The company has not published enough task-generalization results to show how much capability it gets from each extra hour.

Q7Is Tesla really getting ready to mass-produce Optimus?

Yes. Tesla is physically preparing to mass-produce Optimus now, and this is the strongest part of the story.

Tesla’s first large-scale Optimus line is replacing the discontinued Model S and Model X lines in Fremont. The company says that initial line is designed for one million robots per year, while a future Texas line is being designed around ten million units of long-term annual capacity.

The Fremont conversion means more than another production forecast. Tesla has sacrificed real automotive floor space, dismantled existing equipment and assigned the building to Optimus. Recent reporting says the old lines were decommissioned in fewer than seven weeks.

But designed capacity is not expected output. A line can be laid out for one million units and produce only a few thousand if suppliers, yields or the robot design remain unstable. Tesla has not disclosed installed tooling, initial takt time, supplier capacity or how many production-intent robots have actually been completed.

Tesla is also leaning on manufacturing expertise outside Fremont. Its China leadership has said the Shanghai operation should play an important role in solving robot mass-production problems. Shanghai has produced vehicles at enormous scale, although that statement also suggests Tesla has not fully solved the move from hand-built Optimus units to repeatable assembly.

The production effort is real and moving quickly. For now, the one-million-unit figure describes the architecture of a factory, not a dependable output forecast.

Q8Why did Tesla miss its first Optimus production target?

Tesla missed its early Optimus production target because the robot entered 2025 with unresolved hardware, supply-chain and organizational problems.

Musk discussed building at least 5,000 Optimus robots during 2025. Tesla reportedly ended up producing fewer than 500, leaving output below one-tenth of that ambition.

Several problems seem to have landed at once.

The robot’s hands and actuators were still being redesigned, making it difficult to lock suppliers and assembly procedures. Chinese restrictions on rare-earth magnet exports then disrupted access to components used in those actuators. Meanwhile, Milan Kovac, the executive leading Optimus engineering, left Tesla in June 2025 for personal reasons, forcing Ashok Elluswamy to take over during a delicate production phase.

No single event explains a tenfold miss. Together, they show that Tesla set a volume target before the robot architecture and supply chain were stable enough to support it.

That should change how we read the new one-million-unit line. Tesla could ramp rapidly once Gen 3 freezes, especially with its manufacturing resources. Its previous target also shows how far executive confidence can run ahead of engineering readiness.

The Optimus program is moving forward. Tesla’s schedule should still be treated as an aspiration until physical output starts matching it.

Q9Can Tesla manufacture Optimus more cheaply than everyone else?

Tesla may eventually become the lowest-cost Western humanoid producer, but there is currently no data proving that it has a cost advantage.

Tesla starts with assets that competitors would struggle to recreate. It already builds batteries, power electronics, computers and complex electromechanical products in large factories. It has supplier relationships, a distribution and service footprint, and its own plants can act as the first customers.

Its planned scale could also change component economics. A supplier will quote very different prices for 5,000 actuators and five million actuators. Tesla can redesign parts around automated assembly and spread software, tooling and engineering costs across a much larger fleet.

Optimus is not simply a small Tesla vehicle with legs, though. Compact actuators, dexterous hands, force sensing, precision gears and repeated physical impacts create manufacturing and maintenance problems Tesla has never solved at this density.

Chinese competitors already benefit from a domestic robotics supply chain. UBTECH began mass-producing and delivering several hundred Walker S2 robots after reporting orders exceeding RMB800 million. That does not prove UBTECH has better AI. It does show Tesla will not be scaling against an empty field.

Tesla has disclosed no bill of materials, assembly time, yield, maintenance cost or present unit cost for Optimus. Any claim that it already holds a cost lead would be made up.

Our current assessment is strong cost potential with zero verifiable cost evidence. Manufacturing scale could become Tesla’s largest advantage, but first Gen 3 has to come off the line repeatedly without expensive manual rework.

Q10Is Tesla Optimus ready for outside customers now?

Tesla Optimus is not ready for outside customers today, and Tesla has not created the basic commercial machinery required to sell it.

There is no public price, order form, customer contract, warranty, maintenance package, safety certification or delivery calendar. Tesla has not announced an external customer running a paid Optimus deployment.

That puts Tesla behind several competitors on commercial validation. Agility Robotics has a multi-year Robots-as-a-Service agreement with GXO. Figure completed an extended deployment at BMW. UBTECH has reported industrial orders and begun delivering hundreds of Walker S2 units.

Tesla’s internal-first approach is still defensible. Selling too early would create customer-support problems and expose immature hardware. Its own factories offer enough repetitive tasks to refine the robot before an external launch.

Internal testing can hide weaknesses too. Tesla can redesign a workstation, tolerate engineering supervision and count downtime as useful development. A paying customer will compare Optimus with conveyor systems, fixed robot arms, mobile robots and human labor on cost and uptime.

The first external Optimus contract will reveal more than another polished demonstration. It will show which tasks Tesla is willing to support, what performance it will guarantee and whether somebody sees enough value to pay.

Until then, Optimus remains an internal Tesla program rather than a sellable industrial product.

Q11Is Tesla Optimus getting better faster now?

Tesla Optimus is currently accelerating as a manufacturing program, while evidence of autonomous work is improving much more slowly.

The industrial side has moved decisively. Tesla ended production of two vehicle models, cleared their Fremont lines, started preparing a million-unit Optimus facility and brought Shanghai manufacturing expertise into the effort. That is a big increase in commitment compared with the pilot-line language used earlier.

The public capability record has not advanced at the same pace. We have seen better movement, more dexterous hands and broader demonstrations, but still no equivalent of Figure’s 1,250-hour BMW record or Agility’s paid GXO deployment.

This is the central tension in the Optimus story today. Tesla is building the production system before showing outsiders a stable productivity baseline for the robot that system will manufacture.

The strategy could look brilliant if Gen 3 works. Tesla would emerge with a functioning robot and a scale advantage that competitors cannot quickly copy. It becomes much riskier if the hands, actuators or autonomy need another deep redesign after factory tooling is installed.

Optimus is clearly progressing, just not evenly. Tesla’s willingness to manufacture the robot is currently more advanced than the robot’s demonstrated ability to earn its place in a factory.

Q12How much of the Tesla Optimus story can we trust?

We can trust that Tesla is making an enormous Optimus investment today. We cannot yet trust its production forecasts or claims of broad usefulness.

The factory evidence is strong. Tesla has publicly documented the Fremont conversion, described its planned capacity in shareholder materials and stopped building Model S and Model X vehicles there. Those actions involve real capital, real floor space and a visible opportunity cost.

The evidence becomes weaker when Tesla discusses performance. Selected videos do not reveal how many attempts failed, whether an operator assisted, how long the behavior ran or how the robot reacted when conditions changed.

Forecasting is the weakest area. Tesla’s early 5,000-unit ambition ended with reported output below 500, and the Gen 3 reveal and production schedule have already moved. The one-million and ten-million figures are best read as long-term factory-design targets, not near-term production guidance.

So our confidence is uneven. High that Tesla is industrializing Optimus. Medium that mobility and manipulation are improving. Low on current productivity, autonomy rates, unit economics and production timing.

Treating every Tesla claim as completely credible or completely fictional misses the story. The project has become real faster than the proof around the product has become transparent.

Q13What is the latest verdict on Tesla Optimus?

As of today, Tesla Optimus is one of the world’s most serious humanoid programs, but Tesla has not yet shown one of the world’s most productive humanoid robots.

Tesla has cleared several major hurdles. It built a credible biped, improved manipulation, created an embodied-data operation and committed a real factory to the program. Few competitors combine AI development, internal deployment sites, capital and manufacturing ambition at the same level.

Tesla is much less advanced in public validation. Figure has released stronger operating statistics. Agility has a clearer commercial deployment. UBTECH has already reported substantial orders and deliveries. Boston Dynamics and Hyundai are preparing their own vertically integrated manufacturing push.

Optimus could still overtake all of them because Tesla is attacking the problem at a scale few companies can match. Today, though, that argument depends on what Tesla might achieve rather than what it has already measured publicly.

Our current tracker label is: advanced production-intent humanoid with exceptional manufacturing backing, limited operating transparency and no proven commercial economics.

The next upgrade should require harder evidence: production-intent Gen 3 units leaving Fremont, a verified fleet count, sustained task-level autonomy, intervention rates, operating hours, hand-lifetime data and at least one task with a measurable cost advantage.

Tesla Optimus deserves close attention now. It has not yet earned the benefit of the doubt.