Every Humanoid Robot Company in 2026: The Complete Guide

The short version: there is no single “leader” across the humanoid robot market in 2026. Agility Robotics has one of the clearest active commercial warehouse deployments through Digit at GXO. Figure has unusually strong production evidence from Figure 02 at BMW and a new Figure 03 logistics pilot. Unitree publishes a real starting price for G1. 1X accepts deposits for NEO. Boston Dynamics, Tesla, Apptronik and others are at different points between productization, evaluation and external deployment. Those are different achievements and should not be collapsed into one ranking.

This guide is the editorial map. The maintained source of truth is TheMimic's 73-profile Robot Reality Index, where each generation has its own page, and the Deployment Ledger, where announcements, evaluations, pilots and active deployments are kept separate.

The seven companies most buyers and builders compare

The table deliberately avoids valuations, projected unit volumes and target prices. Those numbers change quickly and often come from executives describing future intent. A useful company guide starts with what a buyer, customer or primary source can verify now.

Figure AI: strongest production history, new generation still in pilot

Figure should be split by generation. Figure 02 is a historical robot with completed production evidence: BMW says the robot supported production of more than 30,000 X3 vehicles during a ten-month 2025 deployment, positioning sheet-metal parts in the body shop. Figure says that fleet has now been retired.

Figure 03 is the current platform. BMW announced in June 2026 that it would begin a new sequencing use case in logistics at Plant Spartanburg. That is meaningful customer confirmation, but it is a new pilot rather than proof that Figure 03 is broadly deployed. The distinction matters: performance data from Figure 02 should not silently become a capability claim for Figure 03.

Primary evidence: BMW's Figure 03 project announcement and Figure's Figure 02 deployment report.

Agility Robotics: the cleanest active warehouse deployment record

Digit is a bipedal logistics robot rather than an attempt to reproduce every aspect of the human form. That narrower design focus is part of the commercial story. GXO says Digit robots are operational in a live warehouse at a SPANX facility, moving totes from mobile robots to conveyors under a multi-year Robots-as-a-Service agreement.

Agility has also announced a commercial agreement with Toyota Motor Manufacturing Canada after a pilot. The agreement is important, but the ledger classifies it as contracted until a later source confirms active operations. Signed intent is evidence; it is not the same evidence as robots doing the work.

Digit's public specifications include a 35 lb carrying capacity and up to four hours of battery life on Agility's official material. GXO's earlier pilot release states a height of 5 ft 9 in and weight of 140 lb. Those values are source-attributed rather than independently benchmarked.

Primary evidence: GXO's commercial deployment announcement and Agility's Toyota agreement.

Boston Dynamics: productized Atlas, announced fleets

Electric Atlas is no longer the hydraulic research machine from the viral videos. Boston Dynamics unveiled a product version in January 2026 and said manufacturing would begin immediately. The company also said that 2026 fleets were scheduled for Hyundai's Robotics Metaplant Application Center and Google DeepMind.

That places Atlas beyond a lab-only research project, but the public evidence still needs careful labels. A scheduled fleet is “announced” until shipment and operation are confirmed. TheMimic does not turn corporate ownership by Hyundai into automatic customer validation, and it does not infer factory scale from a product launch.

Primary evidence: Boston Dynamics' 2026 Atlas product announcement.

Apptronik: Apollo 2 is current; external evidence belongs to the earlier generation

Apollo 2 is Apptronik's current humanoid platform as of June 2026, offered in bipedal and wheeled-base configurations for real-world task-data collection. Apptronik says fleets are active across its Robot Park network and at unnamed customer or partner sites. Because those customers are not named in the source, the ledger keeps that record at vendor-confirmed pilot evidence.

The original Apollo is the generation with published launch specifications: 5 ft 8 in height, 160 lb weight, 55 lb lift capacity and four-hour runtime per swappable battery pack.

Mercedes-Benz provides the strongest external evidence. It says Apollo robots have collected data in a production environment at the Digital Factory Campus in Berlin, with employees transferring process knowledge through teleoperation and augmented reality. Mercedes describes work toward autonomous operations and possible intralogistics tasks. That supports an evaluation classification, not a claim of scaled autonomous production.

GXO and Apptronik have also described a laboratory evaluation before any distribution-center deployment. Again, the order of events matters: lab evaluation first, facility deployment only once the system is ready.

Primary evidence: Apptronik's Apollo 2 and Robot Park announcement, Mercedes-Benz's original Apollo project page and Apptronik's original Apollo launch.

1X Technologies: a real order path with explicit early-access limits

NEO is the clearest home-focused product in this group. The official site accepts a $200 deposit and publishes unusually detailed hardware specifications: 5 ft 6 in height, 66 lb weight, an 18 lb arm payload, 842 Wh battery and four-hour stated runtime.

The more important detail is operational. 1X says early owners receive basic autonomy and can schedule a 1X Expert to supervise unfamiliar chores. That is a concrete product model, but it also means buyers should not read the marketing demos as proof of unrestricted household autonomy. The correct status is pre-order, not deployed consumer product.

Primary evidence: 1X's NEO product and order page.

Unitree: public prices make the commercial claim testable

Unitree's G1 and H1 are now separate records. The G1 page lists a starting price of $13,500 before tax and shipping, about 35 kg weight with battery, about two hours of battery life and different joint configurations across the base and EDU versions. Unitree also warns that the humanoid industry remains early and that some functions are still being developed.

That combination is useful: a buyer can see a commercial path and the company's caveat. Public pricing does not prove a customer deployment, and G1's price or specifications should not be attached to H1. The old combined treatment obscured both points.

Primary evidence: Unitree G1 official product page.

Tesla Optimus: manufacturing potential is not deployment evidence

Optimus remains one of the most watched humanoid programs because Tesla has manufacturing infrastructure, AI expertise and a large internal factory environment. None of that is the same as an external customer deployment.

The current profile therefore stays at prototype. It links to Tesla's official materials, records claimed capabilities conservatively and leaves public price, payload and battery fields unknown where current product documentation does not support them. Executive target prices and volume forecasts belong in dated analysis, not as product facts in a maintained directory.

The broader field: industrial, research and social humanoids

The market is much larger than seven companies. The live index includes industrial systems such as HMND 01, Walker S2, Agile ONE and GR-2; research platforms such as TALOS, iCub and ARTEMIS; and social or service robots such as Ameca, Pepper and Mirokaï.

These categories should not be ranked on one scale. A small research platform with an open SDK can be more useful to a university than a full-size factory prototype. A wheeled social robot can be commercially deployed without solving bipedal locomotion. “Humanoid” describes a broad design space, not one buyer requirement.

How to compare humanoid robot companies without getting fooled

Start with four questions:

1. Which generation is being discussed? A deployment by a retired model does not prove the successor has identical reliability.

2. Who confirms the work? Customer documentation is stronger than a vendor-only case study. Both can be useful when labeled.

3. What stage is the relationship? Announcement, contract, evaluation, pilot, active deployment and completed deployment are different states.

4. Which fields are actually public? A public price, payload or runtime should link to the exact source. Missing data should remain missing.

That is the logic behind TheMimic's evidence methodology. It avoids a composite “reality score” because the inputs are not commensurable. A robot can have strong locomotion evidence and weak manipulation evidence, or excellent customer validation for one narrow task and no proof of generality.

Verdict

The humanoid robot race in 2026 is real, but uneven. Agility and Figure have the strongest records in this comparison for documented customer work. Apptronik has credible production-environment evaluation. Boston Dynamics has a productized platform and announced 2026 fleets. Unitree has the most transparent entry pricing among the major full-body platforms tracked here. 1X has turned a home-robot concept into a concrete pre-order proposition, while openly retaining supervised support for early capabilities. Tesla still carries exceptional manufacturing potential without equivalent external deployment evidence.

The useful question is no longer “which demo looks best?” It is “which claim, for which generation, in which environment, is supported by whom?” Use the Robot Reality Index to compare the profiles, and the Deployment Ledger to inspect the evidence behind real-world work.

Published by TheMimic. Robot profiles and deployment records are maintained separately from editorial analysis.